Use embedded loop tags as the first choice and add a lossless trim command

[Xehanort88]

Depends on #76. The first two commits come from that PR (the test suite and bug fixes); only the last commit, Use embedded loop tags as the first choice and add a lossless trim command, is new here. Once #76 is merged, this PR will show only that commit.

Summary

Two related features for tracks that already have loop metadata, which is common for game music.

Embedded loop tags as the first choice

If a file already has loop tags (LOOP_START/LOOP_END, LOOPSTART/LOOPLENGTH, or any other name recognized by the tag auto-detection from #66), those loop points now come first in every command (play, export-points, tag, split-audio, extend, trim), ahead of the detected ones.

• Detection still runs, and its results are listed after the tagged loop, so -i still lets you pick another one. The tagged loop shows as "from tags" in the interactive table, and play reports it as such.
• If detection finds no loop, the tagged points are still returned instead of raising LoopNotFoundError.
• Tags are ignored when --approx-loop-position is given (the user asked for loops near a specific position) and when they fall outside the audio (with a warning).
• New --ignore-tags option to skip them. From Python: find_loop_pairs(use_embedded_tags=False) and read_embedded_loop_pair().
• LoopPair gains a from_metadata field (defaults to False, so existing code is unaffected).

trim command: cut after the loop end without re-encoding

pymusiclooper -i trim --path "TRACK.ogg" --keep-after 500

Writes TRACK-trimmed.ext, keeping everything up to the loop end plus --keep-after N samples (default 0), with the original format, bit depth and tags.

Format	How	Result
WAV (PCM 8–32-bit, float, double)	Samples read and written in their native format	Bit-exact
FLAC	Same; the compressed bytes differ, but the audio does not	Bit-exact decoded audio
Ogg Vorbis	Cut at the container level, with no decoding or re-encoding (new pymusiclooper/ogg.py)	Bit-exact decoded audio, ending on the exact sample
Anything else (MP3, Opus, µ-law WAV, …)	Rejected with an explicit error, since trimming would require lossy re-encoding	—

How the Ogg Vorbis cut works: pages before the cut are copied byte for byte. On the last kept page, the audio packets after the one containing the end sample are dropped, then its granule position is lowered to the exact end sample and the page is flagged as end of stream. The Vorbis spec (section A.2) defines this as how to end a stream on a sample that isn't on a block boundary. Dropping the extra packets keeps the trimmed samples within the final packet, the same layout encoders produce. That matters because FFmpeg only trims within the last packet. Packet durations come from the mode block flags at the end of the setup header, found by scanning it backwards the same way FFmpeg's vorbis_parser.c and liboggz do, and they're checked against the page granule positions. If they don't add up, the cut falls back to page granularity, which is still valid per the spec.

extend's existing tag-copying code is moved into a shared _copy_tags helper, which trim reuses.

Test plan

• uv run pytest: 72 passed (20 new tests covering embedded tags, bit-exact WAV/FLAC/Ogg trims, tag preservation, the Ogg packet durations and fallback, and rejection of unsupported formats)
• Optional tests/test_ogg_ffmpeg.py decodes trimmed Ogg files with FFmpeg as an independent decoder. It uses PML_FFMPEG, tools/ffmpeg/ or PATH, and skips when FFmpeg is missing or too old to honor Vorbis end trimming. It passes with FFmpeg 7.0.2 and 9.0.2.
• Trimmed real-world Ogg Vorbis files (mono to 6 channels, 16–48 kHz) at many cut points. Decoded output is exact with libvorbis and FFmpeg 7.0.2/9.0.2; with FFmpeg only for cuts past the first audio page (~0.1–0.2 s), which is far earlier than any loop end.