fix: declare regex as a core dependency (CitationMixin crashes with ModuleNotFoundError)

[chuenchen309]

Problem

CitationMixin._get_span (instructor/v2/dsl/citation.py:81) does import regex at call time, but regex is not declared anywhere in pyproject.toml's core dependencies or any [project.optional-dependencies] extra (checked all ~25 extras — none of them pull it in).

A stock pip install instructor therefore has CitationMixin — a documented public feature — crash with ModuleNotFoundError the moment validate_sources() runs. The feature is entirely broken on a clean install.

This isn't hypothetical: the repo's own existing test tests/coverage/test_dsl_small_coverage.py::test_citation_keeps_quotes_without_context_and_recovers_fuzzy_quotes already exercises this exact path and fails on a fresh checkout without regex manually installed.

Fix

Add "regex>=2023.0.0" to [project.dependencies] and update uv.lock accordingly.

Not a duplicate of #2430

#2430 ("fix(citations): escape substring quotes before fuzzy span matching", open) fixes a different bug — regex.error on quotes containing metacharacters — and does not touch pyproject.toml. Its own new test uses pytest.importorskip("regex"), i.e. it works around this exact missing-dependency issue rather than fixing it. Both fixes are needed independently.

Testing

• The previously-failing test now passes: 16/16 in test_dsl_small_coverage.py.
• Full tests/dsl/ + test_dsl_small_coverage.py: 88 passed, 2 pre-existing skips (unrelated to this change).
• ruff check clean.

AI-Generated disclosure

Found via an AI-assisted code review pass (Claude Code) over instructor/v2/dsl/. I personally reproduced the ModuleNotFoundError against the repo's own pre-existing test, verified the fix, checked for overlap with #2430, and ran the relevant test suites before submitting.

[ErenAta16]

Checked where regex is actually used: it's a local import regex inside CitationMixin's fuzzy-match method in instructor/v2/dsl/citation.py, not a module-level import, so strictly only the fuzzy-citation path needs it, not every CitationMixin user. Declaring it as a core dependency rather than gating it behind a lazy import or an extras group is still the right call here: regex is small and has no heavy transitive dependencies, and the alternative (lazy-guarded import with a custom error message) would be more code for a package this lightweight. Straightforward fix for a real crash.