feat: add FLAC as a Stream-to-Stream ConvertFile/CutFile destination

[cloud-hai-vo]

Summary

• Continues docs/feature-roadmap.md item 1's own "destination-Stream" half, which the owner's own cron instructions called out by name as an example next step. WAV/AIFF/AU were already done (none of the three need Stream.CanSeek); FLAC is the first destination here that genuinely does.
• FlacEncoder.OpenSession(Stream destStream, ...) -- a new overload, mirroring WavWriter/AiffWriter/AuWriter's own Create(Stream, ...) overloads. Requires destStream.CanSeek (throws ArgumentException up front if not) because FlacEncoderSession.Finish() still seeks back to a fixed byte offset to patch STREAMINFO's provisional totalSamples/MD5/block-size-range fields once their real values are known -- exactly the way the path-based overload's own FileStream destination already does. Does not dispose the caller's stream.
• FlacEncoderSession's own destination field was widened from FileStream to Stream, plus a leaveOpen flag (mirroring WavWriter's own _leaveOpen convention) so Dispose() only closes a stream this type actually owns.
• Channel/bits-per-sample/tag validation was factored out of the existing path-based OpenSession into a shared private helper (ValidateAndGetCommentListBytes) so the two overloads can never drift apart on what's a valid request.
• Wired into AudioCutter.Stream.cs's OpenSinkStream as the AudioFormat.Flac case -- unlike WAV/AIFF/AU, no DeferredFixedHeaderSink wrapper is needed at all, since FLAC's own header-patch-after-the-fact design never needed an exact frame count up front in the first place (matching the existing path-based AudioCutter.cs dispatch for .flac, which already calls FlacEncoder.OpenSession directly with no such wrapper).
• Found and fixed two pre-existing, now-stale tests during this change: NativeEncoderTest.ConvertFile_Stream_WithUnsupportedDestFormat_Should_Throw and AudioCutterStreamTest.Convert_Stream_WithUnsupportedDestFormat_Should_Throw both used AudioFormat.Flac as their "still unsupported" example destination -- now true, so both were switched to AudioFormat.Wma (still genuinely unsupported as a destination).
• Found and fixed stale documentation unrelated to this change but touched by it: CLAUDE.md's own AudioCutter.Stream.cs row still said "Destination: WAV only so far", which had already drifted out of sync with reality once AIFF/AU landed in an earlier PR; corrected while updating the same line for FLAC.
• Docs synced: CLAUDE.md, README.md, llms.txt, llms-full.txt, docs/feature-roadmap.md item 1's own status note.

Test plan

• New FlacEncoderTest cases: null stream, non-seekable stream (throws with "seekable" in the message), zero channels, bits-per-sample out of range (0/33), invalid tag field name (throws before a single byte reaches the caller's stream), exact round-trip via a MemoryStream destination, and Dispose() not disposing the caller's stream.
• New AudioCutterStreamTest cases: WAV source → FLAC Stream destination round-trip (exact samples), non-seekable FLAC destination throws through the full AudioCutter.Convert call, and a ResamplingTransform pipeline into a FLAC Stream destination writes the correct total frame count.
• New shared test utility NonSeekableWriteStream (the write-side counterpart to the existing NonSeekableStream).
• Full suite (dotnet test, net10.0 locally): 1729 passed (1718 baseline + 11 new), 1 pre-existing unrelated skip, 0 failed.
• Full solution build: 0 warnings, 0 errors.

🤖 Generated with Claude Code