Add capability-driven audio write and conversion pipeline - #125

Summary

Follow-up to merged #106. This PR turns the newly available managed audio encoders into a reusable container write/transcode pipeline instead of adding pairwise converters.

Common pipeline

• container-neutral PCM and encoded-access-unit contracts in Compression.Registry
• Compression.Lib.AudioConversionOperation routes conversions in least-destructive order:
  1. same-format byte-exact passthrough
  2. packet-preserving demux/remux
  3. canonical PCM decode/encode
  4. existing per-channel WAV pseudo-archive bridge
• existing formats that expose channel WAVs automatically become transcode sources

Implemented write participants so far

• AAC/ADTS: PCM encode/decode, archive create, raw AAC access-unit demux + AudioSpecificConfig
• MP4/M4A: audio-only AAC-LC muxer with ISO BMFF timing/sample tables and direct AAC access-unit remux
• FLAC: canonical PCM source/target and channel-WAV create surface
• Ogg: managed Vorbis/Opus encode/create plus canonical PCM decode

Tests

• byte-exact same-format passthrough
• AAC -> M4A preserves every compressed AAC access unit
• WAV -> FLAC lossless round trip
• WAV -> Ogg Vorbis
• WAV -> Ogg Opus
• WAV -> M4A encode/mux
• explicit unsupported codec/container failure

Next in this PR

Extend the same interfaces to MP3, WavPack, WAV/AIFC/AU compressed modes and then game/multi-stream containers where a real semantic writer exists. Capability/documentation states will only be promoted after writer coverage passes.