parse_encoding rejects a UTF-8 BOM when the encoding declaration uses UTF-8, utf_8, UTF_8 or utf-8-sig, although Python accepts these declarations. Consequently, message extraction fails before tokenization on valid source files.

Normalize case and underscores only for the BOM compatibility check, accepting utf-8 and utf-8-* as Python does. Keep the existing utf-8 return value, preserve the original declaration in error messages and leave non-BOM behavior unchanged. General codec lookup is intentionally avoided: Python rejects BOM declarations such as utf8 and cp65001 even though they identify UTF-8 codecs.

Tests cover both declaration lines, valid aliases, incompatible declarations and stream-position restoration on success and failure. Each accepted example is first compiled by Python to check that it is valid source.

Validation on Windows Python 3.12.14:

• Original utility/extraction baseline: 24 passed.
• Regression tests before the fix: 8 failed, 8 passed.
• Fixed utility/extraction selection: 37 passed, independently rerun by a separate reviewer.
• Full suite with checksum-verified CLDR 48.2 imported by the repository script: 7841 passed, 10 skipped, 2 xfailed. One unrelated pytest deprecation warning.
• A direct extract_python smoke check now extracts a message from a BOM-prefixed source using UTF_8.
• Ruff and every applicable repository-pinned pre-commit hook pass on the two changed files.

Other Python versions and operating systems have not been tested locally.

AI assistance: this is an autonomous, user-authorized Codex contribution. Codex investigated, implemented and tested the change; a separate agent reviewed it and reran the targeted tests. No manual human review or production incident is claimed.