ENH: Migrate numpydoc validation to sphinx

[tuhinsharma121]

Summary

• Removes the standalone scripts/validate_docstrings.py (480 lines) and its test suite scripts/tests/test_validate_docstrings.py by integrating all pandas-specific docstring checks directly into the Sphinx build via a monkey-patch of numpydoc.validate in doc/source/conf.py.
• Enables numpydoc_validation_checks = {"all"} with an explicit numpydoc_validation_exclude set of 171 fully-qualified object names (replacing the previous 25 regex wildcards), so every public API object is validated during the documentation build.
• Removes the separate ci/code_checks.sh docstrings CI step and its corresponding GitHub Actions workflow entry, since validation is now handled entirely by Sphinx.

Motivation

(following up on #64565 (comment))

Previously, pandas maintained two independent validation paths: numpydoc's built-in Sphinx checks and a custom standalone script (scripts/validate_docstrings.py) that enforced four pandas-specific rules (GL04, PD01, SA05, EX04). This had several downsides:

1. Duplicate infrastructure -- two separate validation tools, each with their own exclusion lists and test suites.
2. Inconsistent results -- a docstring could pass the standalone script but fail during the Sphinx build, or vice versa.
3. Maintenance burden -- nearly 1,000 lines of custom validation code and tests to maintain.

By monkey-patching numpydoc.validate.validate in conf.py, all checks (standard numpydoc + pandas-specific) now run in a single pass during the Sphinx build. Contributors only need one command to validate a docstring.

Changes

doc/source/conf.py

• Added monkey-patch that injects four custom error codes (GL04, PD01, SA05, EX04) into numpydoc.validate.ERROR_MSGS and wraps the validate function.
• Added numpydoc_validation_checks = {"all"} to enable comprehensive validation.
• Added numpydoc_validation_exclude with 171 explicit fully-qualified entries replacing the previous regex wildcards for objects that legitimately cannot have numpydoc-compliant docstrings (e.g., Jinja2 template attributes) or are missed out (e.g. offset .base/.rollback/.rollforward methods, ExtensionDtype stubs).

scripts/validate_docstrings.py (deleted)

scripts/tests/test_validate_docstrings.py (deleted)

ci/code_checks.sh

• Removed the ### DOCSTRINGS ### section that invoked scripts/validate_docstrings.py.
• Removed docstrings from the usage instructions and argument validation.
• Removed the now-unused BASE_DIR variable.

.github/workflows/code-checks.yml

• Removed the Run docstring validation step (ci/code_checks.sh docstrings).

doc/source/development/contributing_documentation.rst

• Replaced references to scripts/validate_docstrings.py with the single recommended command for validating a docstring:
  python doc/make.py --warnings-are-errors --no-browser --single pandas.DataFrame.mean
• Updated the description of docstring validation to explain that all checks now run during the Sphinx build.

doc/source/development/contributing_codebase.rst

• Updated the description of ci/code_checks.sh to reflect that standard numpydoc validation is now enforced during the Sphinx build.

Test plan

• Run python doc/make.py --warnings-are-errors --no-browser --single pandas.DataFrame.mean and verify it completes without errors.
• Deliberately introduce a SA05 violation (e.g., add pandas.Series.rename to a See Also section) and verify the single-page build catches it.
• Run ./ci/code_checks.sh and verify all remaining checks (code, doctests, single-docs, notebooks) pass without the removed docstrings step.
• Run a full doc build (cd doc && python make.py --num-jobs 1 html) and verify no regressions in validation warnings.

[mroeschke]

Thanks! Generally looking good