feat(torch): add generated operator bases

[voltjia]

Summary

• Add generated C++ operator base headers under src/base/, regenerated from the current PyTorch codegen output merged in PR feat(torch): expose optional codegen parameters #619.
• Keep the generated base files flat under src/base/; no generator, test, wrapper, build-system, or CI files are changed in this PR.
• Omit generated bases whose ATen schemas are known to vary across installed PyTorch builds, so they can remain generated by the local codegen environment instead of frozen as stable public bases.

Motivation

The PyTorch codegen work needs a checked-in operator-base layer that matches the current generator behavior, including optional-parameter overload support from PR #619. This PR contains only the generated public base headers, making the downstream base layer reviewable separately from the generator changes.

Closes # N/A — no dedicated issue.

Type of Change

• feat — new feature / new operator / new platform.
• N/A — fix — bug fix.
• N/A — perf — performance improvement (no behavioral change).
• N/A — refactor — code restructuring without behavior change.
• N/A — test — adding or fixing tests only.
• N/A — docs — documentation only.
• N/A — build / ci — build system or CI configuration.
• N/A — chore — tooling, formatting, or other non-code changes.
• N/A — Breaking change.

Platforms Affected

• CPU (WITH_CPU)
• NVIDIA (WITH_NVIDIA)
• Iluvatar (WITH_ILUVATAR)
• MetaX (WITH_METAX)
• Cambricon (WITH_CAMBRICON)
• Moore (WITH_MOORE)
• Ascend (WITH_ASCEND)
• PyTorch C++ bindings (WITH_TORCH)
• N/A — Build system / CMake / CI; no build-system or CI files are changed.
• Python bindings / user-facing API

Test Results on Supported Platforms

All rows used a full bare python3 -m pytest -v run, without tests/, --devices, or -n. Each build regenerated PyTorch operator sources first, installed with WITH_TORCH=ON, and smoke-checked representative generated PyTorch operators after install. Build times are from the pip install phase recorded by the local validation runner; pytest times are from the timed pytest command; total time is build + pytest.

Platform	Built	pytest Result	Build	Pytest	Total	Notes / Hardware
NVIDIA	Yes	9279 passed, 8565 skipped	1126s	396s	1522s	Full bare pytest. PyTorch backend compiled and generated torch-op tests were included.
Iluvatar	Yes	7777 passed, 8549 skipped	821s	639s	1460s	Full bare pytest. PyTorch backend compiled and generated torch-op tests were included.
MetaX	Yes	8771 passed, 7555 skipped	1448s	436s	1884s	Full bare pytest. PyTorch backend compiled and generated torch-op tests were included.
Cambricon	Yes	5974 passed, 9968 skipped	2308s	999s	3307s	Full bare pytest. PyTorch backend compiled and generated torch-op tests were included.
Moore	Yes	8537 passed, 7807 skipped	2322s	671s	2993s	Full bare pytest. PyTorch backend compiled and generated torch-op tests were included.
Ascend	Yes	7454 passed, 8830 skipped	1125s	615s	1740s	Full bare pytest. PyTorch backend compiled and generated torch-op tests were included.

Full `pytest` output (optional)

NVIDIA:    9279 passed, 8565 skipped in 390.66s (0:06:30)
Iluvatar:  7777 passed, 8549 skipped in 635.02s (0:10:35)
MetaX:     8771 passed, 7555 skipped in 418.39s (0:06:58)
Cambricon: 5974 passed, 9968 skipped in 990.75s (0:16:30)
Moore:     8537 passed, 7807 skipped in 662.66s (0:11:02)
Ascend:    7454 passed, 8830 skipped in 597.06s (0:09:57)

The test counts are expected to match the PyTorch codegen coverage from PR #619 because this PR only checks in generated base headers from that generator. The only observed difference from the latest PR #619 table is on Ascend: one tests/test_torch_ops.py inner case is skipped instead of passed:

tests/test_torch_ops.py::test_op[npu-dtype1-0.01-0.01-13x4-inner]

The generated inner base, binding metadata, and PyTorch backend source are identical between PR #619's generated output and this PR's checked-in base. Ascend still builds successfully, smoke checks show the PyTorch slot active for Ascend, and the full pytest run exits successfully. This is recorded as a non-blocking skip-count drift rather than a build or execution regression.

Benchmark / Performance Impact

N/A — this PR checks in generated base headers only. The table above records build and test wall time for each platform.

Notes for Reviewers

This PR is rebased on the latest master, after PR #619 was merged. The generated base files are intentionally checked in as generator output. File paths are kept flat under src/base/.

The generated bases intentionally omit src/base/all.h, src/base/any.h, and src/base/internal_scaled_mm.h in this PR because their ATen schemas vary across installed PyTorch builds; those forms are better regenerated by the local codegen environment instead of frozen as stable public bases.

---

Checklist

Title, Branch, and Commits

• PR title follows Conventional Commits (e.g. feat(nvidia): …, fix(cuda/gemm): …).
• Branch name follows <type>/xxx-yyyy-zzzz where <type> matches the PR title's Conventional Commits type and words are joined with hyphens (see CONTRIBUTING.md §Branches).
• Each commit message follows Conventional Commits.
• Small PR is a single squashable commit; or, for a large PR, every commit is meaningful, well-formed, and independently reviewable (see CONTRIBUTING.md §Pull Requests).
• No stray merge commits from master — the branch is rebased cleanly on top of current master.
• No fixup! / squash! / wip commits remain.

Scope and Design

• Changes are minimal — nothing unrelated to the stated motivation was added (CONTRIBUTING.md §Code/General).
• No dead code, commented-out blocks, debug prints, printf/std::cout/print(...) left behind, or TODO without an owner and issue link.
• No unrelated formatting churn that would obscure the diff.
• Public API changes are intentional, documented in this PR, and reflected in affected callers/tests.

General Code Hygiene

• The code is self-explanatory; comments were added only where the why is non-obvious (CONTRIBUTING.md §Code/General).
• Every modified or added file ends with a single trailing newline (CONTRIBUTING.md §Code/General).
• No trailing whitespace, tab/space mixing, or stray BOMs.
• Identifiers in comments and error messages are wrapped in backticks (e.g. the `seqlens_k` tensor) (CONTRIBUTING.md §Code/General).
• All comments and error messages are in English (CONTRIBUTING.md §Code/General).
• Comments and error messages are complete sentences — capitalized first letter, terminal punctuation — unless the language/framework convention says otherwise (CONTRIBUTING.md §Code/General; §Python).

C++ Specific

• Code follows the Google C++ Style Guide strictly.
• clang-format --dry-run --Werror passes on all modified src/base/*.h files.
• clang-tidy concerns (per .clang-tidy) have been reviewed — no new warnings beyond the existing baseline.
• Operator parameter order is inputs first, outputs last; attributes are between inputs and outputs; naming follows PyTorch → ONNX → CUDA API precedence (CONTRIBUTING.md §C++).
• No exceptions are thrown. Error paths use assert with messages that include at least __FILE__, __LINE__, and __func__ (CONTRIBUTING.md §C++).
• N/A — No new C++ error or warning message was added.
• N/A — Kernel files are named correctly; this PR adds operator bases, not kernels.
• N/A — Kernel and kernel launcher separation is unchanged; this PR adds operator bases, not kernels.
• Constructor initializer list order matches member declaration order (CONTRIBUTING.md §C++).
• Exactly one blank line between classes, between classes and functions, and between functions (CONTRIBUTING.md §C++).
• Exactly one blank line between members within a class (CONTRIBUTING.md §C++).
• Exactly one blank line before and after the contents of a namespace (CONTRIBUTING.md §C++).
• New operators added via src/base/<op>.h (inheriting Operator<Op>) with generated PyTorch backends provided by PR feat(torch): expose optional codegen parameters #619 (CONTRIBUTING.md §Adding an Operator).
• No raw new/delete; RAII / smart pointers / existing allocators are used.

Python Specific

• N/A — This PR does not modify Python files.

Testing

• pytest was run locally on every supported platform that this PR can affect, and the results are recorded in the "Test Results" table above (CONTRIBUTING.md §Pull Requests).
• N/A — Every supported platform was tested.
• New functionality is covered by PR feat(torch): expose optional codegen parameters #619's generated PyTorch operator test harness and the all-platform full pytest runs recorded above.
• N/A — This PR does not add Python tests.
• N/A — This PR does not add flaky parallel-only tests.
• N/A — This is not a bug-fix-only PR.

Build, CI, and Tooling

• The project builds cleanly from a fresh directory on every supported platform listed above.
• compile_commands.json still regenerates through the existing CMake/scikit-build configuration path.
• N/A — No new backend / device was added.
• Only one CUDA-like GPU backend is selectable at a time — the existing mutual-exclusion check in CMakeLists.txt is not broken.
• Both CI workflows (clang-format.yml, ruff.yml) are expected to remain green; this PR only changes generated C++ headers under src/base/.
• N/A — No new runtime dependency was added.

Documentation

• N/A — No user workflow, build flag, or developer workflow documentation changed.
• New generated operator bases are documented through their checked-in header signatures.
• N/A — No user-visible breaking change is introduced.

Security and Safety

• No secrets, access tokens, internal URLs, customer data, IP addresses, or personal hardware identifiers have been committed or included in this PR description.
• N/A — No third-party code was added.
• No unsafe pointer arithmetic, uninitialized reads, or missing bounds checks were introduced.