v0.0.8

Other

• parse plst (Playlist) chunk per Microsoft RIFF MCI §3
• parse smpl (Sampler) + inst (Instrument) chunks
• parse cue + LIST adtl per Microsoft RIFF MCI §3
• parse bext Broadcast Audio Extension chunk (EBU Tech 3285)
• fix intra-doc links to WavMuxOptions / WavDemuxer methods
• A-law/μ-law tag 6/7 + WAVEFORMATEXTENSIBLE end-to-end

Added

• WAV demuxer parses the plst (Playlist) chunk per
  docs/container/riff/metadata/microsoft-riffmci.pdf §3 ("Playlist
  Chunk"). The 4-byte dwSegments count is followed by N × 12-byte
  <play-segment> records (dwName, dwLength, dwLoops). Each
  segment surfaces under wav:plst.count plus per-segment
  wav:plst.<n>.cue_id / .length / .loops keys. Unlike the cue
  chunk which keys by dwName, the playlist is keyed by zero-based
  segment position because the spec explicitly allows a single cue
  id to appear in multiple segments (a cue replayed N times = N
  segments with identical dwName), so a dwName-indexed key would
  collide. A dwSegments count that exceeds the chunk body is
  clamped to the records that actually fit; bodies shorter than the
  4-byte segment-count header are treated as opaque and skipped.
• WAV demuxer parses the smpl (Sampler) and inst (Instrument)
  chunks per docs/container/riff/metadata/exiftool-riff-tags.html §
  "RIFF Sampler Tags" / "RIFF Instrument Tags" and summarised in
  docs/container/riff/metadata/README.md § "Sampler / Instrument
  chunks". smpl is a 36-byte fixed header (Manufacturer,
  Product, SamplePeriod, MIDIUnityNote, MIDIPitchFraction,
  SMPTEFormat, SMPTEOffset, cSampleLoops, cbSamplerData)
  followed by N × 24-byte loop records (dwCuePointID, dwType,
  dwStart, dwEnd, dwFraction, dwPlayCount). inst is a
  7-byte fixed struct (UnshiftedNote, FineTune, Gain, LowNote,
  HighNote, LowVelocity, HighVelocity). Both surface through
  Demuxer::metadata under wav:smpl.* / wav:inst.* keys; the
  SMPTEOffset DWORD is rendered as canonical HH:MM:SS:FF, and
  FineTune / Gain are decoded as signed i8 (so -3 cents shows
  as -3, not 253). A cSampleLoops count exceeding what the chunk
  body actually carries is clamped to the records that fit; bodies
  shorter than the 36-byte (smpl) or 7-byte (inst) fixed header
  are treated as opaque and skipped.
• WAV demuxer parses the cue chunk and the LIST adtl (Associated
  Data List) sub-chunks per
  docs/container/riff/metadata/microsoft-riffmci.pdf §3
  ("Cue-Points Chunk" + "Playlist Chunk" + "Associated Data Chunk").
  The cue-point table surfaces under wav:cue.count plus per-point
  wav:cue.<dwName>.position / .fcc_chunk / .chunk_start /
  .block_start / .sample_offset (all values as decimal strings;
  fcc_chunk rendered as a 4-byte ASCII FOURCC when printable). The
  LIST adtl body's labl / note / ltxt sub-chunks surface
  under wav:adtl.labl.<dwName> / wav:adtl.note.<dwName> /
  wav:adtl.ltxt.<dwName>.{length,purpose,text}. file sub-chunks
  (embedded media files) are skipped — their bytes do not fit the
  string-typed metadata API. A dwCuePoints count that exceeds the
  chunk body is clamped to the records that actually fit so a writer
  that lies about the count cannot panic the parser; labl / note
  shorter than the 4-byte dwName header and ltxt shorter than the
  20-byte fixed header are treated as opaque and skipped.
• WAV demuxer parses the bext Broadcast Audio Extension chunk per
  docs/container/riff/metadata/ebu-tech3285-bwf.pdf (EBU Tech 3285 v2
  §2.3 BROADCAST_EXT). The 602-byte fixed struct plus variable
  CodingHistory is surfaced through Demuxer::metadata under
  wav:bext.description / .originator / .originator_reference /
  .origination_date / .origination_time / .time_reference (64-bit
  sample count reassembled from the low/high DWORDs) / .version /
  .umid (hex-encoded SMPTE-330M UMID, v1+ only when non-zero) /
  .coding_history. For BWF v2 the five loudness WORDs are decoded
  from the §2.4 round(100 × value) fixed-point to two decimals under
  .loudness_value / .loudness_range / .max_true_peak_level /
  .max_momentary_loudness / .max_short_term_loudness. Loudness and
  UMID keys are omitted for v0/v1 streams that leave those fields zero;
  a bext chunk shorter than 602 bytes is skipped as opaque.
• WAV demuxer dispatches WAVE_FORMAT_ALAW (0x0006) and
  WAVE_FORMAT_MULAW (0x0007) streams to the pcm_alaw / pcm_mulaw
  codecs (host runtime applies G.711 decode through oxideav-g711).
  Decoded SampleFormat hint surfaces as S16; bit_rate reflects
  the on-wire 8-bit rate (not the post-decode S16 rate).
• WAV WAVE_FORMAT_EXTENSIBLE (0xFFFE) end-to-end handling per
  docs/container/riff/waveformatextensible/README.md: the 22-byte
  extension's wValidBitsPerSample, dwChannelMask and SubFormat
  GUID are parsed and exposed both through typed accessors on
  WavDemuxer (format_tag, valid_bits_per_sample, channel_mask,
  subformat, subformat_text) and through Demuxer::metadata
  under the keys wav:fmt.valid_bits_per_sample /
  wav:fmt.channel_mask / wav:fmt.subformat. Well-known
  KSDATAFORMAT_SUBTYPE_* GUIDs (PCM, IEEE_FLOAT, ALAW, MULAW)
  resolve to the legacy codec ids; unknown GUIDs synthesise a
  wav:guid_<canonical-text> id so downstream make_decoder
  failures name the actual GUID rather than the opaque 0xFFFE tag.
• WAV muxer open_muxer_with + WavMuxOptions::with_extensible(mask)
  emits a 40-byte WAVEFORMATEXTENSIBLE fmt chunk with caller-
  supplied dwChannelMask. wValidBitsPerSample defaults to the
  container wBitsPerSample and SubFormat defaults to the well-known
  GUID for the codec; both can be overridden via
  with_valid_bits_per_sample / with_subformat.
• New public constants WAVE_FORMAT_PCM / WAVE_FORMAT_IEEE_FLOAT /
  WAVE_FORMAT_ALAW / WAVE_FORMAT_MULAW / WAVE_FORMAT_EXTENSIBLE
  in the wav module for muxer callers.

Fixed

• WAV demuxer rejects WAVE_FORMAT_EXTENSIBLE streams whose
  cbSize < 22 (the spec mandates a 22-byte extension; previously
  the demuxer silently dropped the extension fields).
• WAV demuxer stamps the on-wire wFormatTag onto
  CodecParameters.tag so consumers can distinguish legacy
  WAVEFORMATEX from EXTENSIBLE round-trips without reparsing.