bug(postprocess): --recode-audio=auto into .ogg writes Opus, but .ogg means Vorbis (Xiph / RFC 7845)

[crippledgeek]

Found while designing the #618 fix (container → default audio encoder). Split out
because it undoes a deliberate existing choice and is a user-visible behaviour change
in its own right, rather than part of #618's catch-all removal.

The defect

audio_encoder_registry::select_audio_encoder_for_container
(crates/rdlp-ffmpeg/src/ffmpeg/audio_encoder_registry.rs:368) overrides
ContainerFormat::Ogg to Opus:

ContainerFormat::Ogg => preferred_audio_encoder("opus")
    .or_else(|| preferred_audio_encoder("vorbis"))
    .unwrap_or("libopus"),

FFmpeg's ogg muxer declares vorbis as its default audio codec (measured:
av_guess_format("x.ogg") → muxer ogg, av_guess_codec(AUDIO) → vorbis).

So --recode-audio=auto targeting .ogg writes Opus inside a file whose extension
signals Vorbis.

Why that's wrong

Xiph.Org — MIME Types and File Extensions:

".ogg applies now for Vorbis I files only."

The wiki adds that .ogg was "more recently also been used for Ogg FLAC and for Theora,
too — these uses are deprecated now in favor of .oga and .ogv respectively", with the
rationale given as backwards-compatibility with hardware players.

[RFC 7845] §9 ("Content Type"):

"The RECOMMENDED filename extension for Ogg Opus files is '.opus'."

Neither source forbids Opus-in-.ogg, so this is a conformance/expectation issue
rather than a spec violation. But hardware players that treat .ogg as Vorbis are
exactly the population the Xiph recommendation exists to protect.

Precedent

This is the same class as #538, where --remux=wmv wrote Title.asf. The resolution
there was that the extension the user asked for wins over internal canonicalisation,
and Wmv/Wma/Asf became distinct variants so the spelling survived to the output
filename.

rdlp already has a separate ContainerFormat::Opus mapping to .opus, so a user who
wants Opus has a correct and unambiguous way to ask for it. The override gives Opus to
users who did not ask for it.

Options

1. Drop the override — Ogg resolves to the muxer default (vorbis). Most
   conformant; matches the bug(postprocess): --remux wmv produces a .asf file, inverting Microsoft's own naming rule #538 precedent. Cost: quality regression vs Opus for anyone
   relying on the current behaviour.
2. Add an Oga variant for .oga (Xiph's extension for non-Vorbis Ogg audio) and
   route Opus-in-Ogg there. More complete, larger change, and .oga has poor real-world
   player support.
3. Keep as-is, documenting the deviation explicitly rather than leaving it implicit.

Leaning toward (1) on the #538 precedent, but it is a judgement call about existing
user-visible behaviour, not a clear-cut bug.

Verified vs not

Verified (measured against the linked FFmpeg build, 2026-07-22):

• ogg muxer declares vorbis; .opus resolves to the opus muxer declaring opus.
• rdlp has no Oga variant; Ogg and Opus are both in is_audio_only().
• The Xiph and RFC 7845 quotes above are verbatim from the cited pages.

NOT verified — do not treat as established:

• What yt-dlp does by default for .ogg targets. A search was inconclusive; read the
  postprocessor source before citing it as prior art either way.
• Whether any specific hardware player actually rejects Opus-in-.ogg. The Xiph
  rationale asserts the compatibility concern; I did not test a device.
• Whether any rdlp user depends on the current Opus behaviour.

Acceptance

• A decision recorded among the three options, with rationale
• If (1): Ogg resolves via the muxer default; test asserts .ogg → vorbis and
  .opus → opus, and that the two do not collapse
• Behaviour change noted in the changelog if user-visible

Refs #618, #538