Validate license files are valid UTF-8 text in bdist_wheel

[CAOShurong]

Fixes #4936

Problem

PEP 639 requires tools to assume license file content is valid UTF-8 encoded text, and states that tools SHOULD validate this and raise an error if it is not. setuptools currently copies whatever files match the license-files patterns — including binary files like BadLicense.doc, LICENSE.pdf or even a stray license.py matched by the default globs (LICEN[CS]E*) — straight into .dist-info/licenses/ with no warning or error (issue #4936).

Fix

Add _validate_license_file_encoding() in setuptools/command/bdist_wheel.py. It runs for every license file just before it is copied into the wheel's licenses/ folder (in egg2dist), and raises InvalidConfigError when UTF-8 decoding fails:

error: license file 'license.pdf' is not valid UTF-8 encoded text.
Per PEP 639, license files must be UTF-8 text; remove it from
`license-files` or exclude it via more specific patterns.

Because the validation lives in bdist_wheel's copy loop, it covers both explicit patterns and the default globs, on every platform, regardless of how the file list was resolved.

Verification (all on real Windows 11)

• New regression test test_binary_license_file_raises: fails on main (binary file silently bundled), passes with this change — RED/GREEN proven.
• Full setuptools/tests/test_bdist_wheel.py: 29 passed, 2 skipped (unchanged from baseline).
• test_wheel.py + config/ suites: 204 passed, 1 xfailed (unchanged).
• ruff check + ruff format --check clean; mypy setuptools/command/bdist_wheel.py: no issues.
• End-to-end repro of the issue scenario via setuptools.build_meta.build_wheel():
  • license.pdf containing invalid UTF-8 bytes + default globs → build now fails with the error above (previously succeeded and shipped the binary).
  • Valid UTF-8 LICENSE.txt → wheel builds normally, license entries present in .dist-info/licenses/.
• POSIX behavior unchanged: the check only triggers on genuine decode failures; all previously-valid builds keep passing (UTF-8 text with BOM-free multi-byte content decodes identically on any platform).

Notes for reviewers

• The PEP says SHOULD (not MUST), so if maintainers prefer a warning first (e.g. SetuptoolsWarning with a deprecation-style due date) I'm happy to convert this — the mechanism would stay identical, only the raise becomes an emit.
• Scope deliberately limited to the wheel-copy path; sdist handling is untouched.

DCO sign-off included in the commit.

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request