v0.12.14 — discover MP3 audio in session directories

Companion release to vezir 0.8.11 (MP3 upload support).

Added

• MP3 session-dir discovery. millet transcribe <session_dir> and _find_session_files (used by label, ingest, enroll, sync) now discover *.mp3 audio, in addition to *.wav and *.ogg. Preference order is WAV → OGG → MP3; the discovered audio is still surfaced under the "wav" key for backward compatibility.
• Decoding already worked for any ffmpeg-readable format (whisperx.load_audio is ffmpeg-backed); only the directory-discovery globs were restricted. An explicit millet transcribe path/to/file.mp3 already worked and is unchanged.

Release note

This is a re-release of the 0.12.13 tag, whose CI build produced the wrong version (the static version in pyproject.toml had not been bumped, only __init__.py), so PyPI rejected it as a duplicate of 0.12.12. No code changes versus 0.12.13.