fix(axbuild): infer diff base for zero since ref

[YanLien]

Problem

When a newly created remote branch triggers a push CI run for the first time, GitHub Actions sets github.event.before to the all-zero SHA. The CI passes that value to cargo xtask clippy --since, so the incremental diff base cannot be resolved and the check falls back to a full workspace run.

Changes

• Teach the axbuild git helper to recognize an all-zero --since ref.
• Infer the incremental diff base from the local git graph by using the parent of the first commit unique to the current branch.
• Reuse the resolved diff base when classifying root Cargo.toml changes, so zero-since runs do not try to read old files from the all-zero ref.
• Add unit coverage for first-push branch behavior, advanced default-branch tips, root Cargo.toml changes, and inference fallback.

[mai-team-app]

Review Summary

PR 做了什么
当新创建的远程分支首次触发 push CI 时，GitHub Actions 会将 github.event.before 设为全零 SHA。旧代码会直接 bail 并退回全量 workspace clippy。本 PR 教 axbuild 的 git helper 识别零 since ref，从本地 git graph 推断增量 diff base，并复用推断出的 base 来正确检测根 Cargo.toml 变更。

实现分析

• changed_paths_since → changed_paths_since_with_base 重构很干净，让 select_incremental_packages 能使用推断出的 diff base 而非原始零 SHA 来检查 root_manifest_change_since。
• infer_zero_since_diff_base 通过 git rev-list --reverse --parents HEAD --not <remote-refs> 找到当前分支第一个独有的 commit，取其 parent 作为 diff base，逻辑正确。
• git_remote_refs_not_at_commit 只查 refs/remotes，在 CI push 场景下足够。
• 错误处理全面，context 信息清晰。

测试覆盖

• zero_since_on_new_branch_resolves_to_first_unique_parent：基本场景
• zero_since_ignores_default_branch_tip_after_branch_fork：dev 在 fork 后有新 commit 的场景
• zero_since_root_manifest_change_uses_inferred_base：端到端测试，验证 select_incremental_packages 在零 since 下正确检测 Cargo.toml 变更
• zero_since_without_remote_refs_returns_error：无 remote ref 时的降级行为

CI 状态
Clippy (run_host) ✅ | Test starry riscv64 qemu (run_container) ✅ | Test arceos aarch64 qemu (run_host) ✅ | Test arceos riscv64 qemu (run_host) ✅ | Test starry self-hosted board orangepi-5-plus (run_host) ✅。skipped jobs 是 run_host/run_container 矩阵互斥的预期行为。

重复分析
无重叠 open PR。#1160 (Wayland)、#1159 (CI dedupe) 与本 PR 改动的 scripts/axbuild/src/support/git.rs 无关。

本地验证

• cargo fmt --check ✅
• cargo clippy --all-features -- -D warnings ✅
• cargo test -- support::git 22/22 通过 ✅（含 4 个新增测试）

结论：代码逻辑正确、测试覆盖充分、CI 全部通过、无阻塞问题。APPROVE。

Powered by mimo-v2.5-pro