Adds E-AC-3 / Dolby Atmos (E-AC-3 JOC) support across the MP4 extractor and muxer.

[rexjin-meta]

Adds E-AC-3 / Dolby Atmos (E-AC-3 JOC) support across the MP4 extractor and muxer.

• New FormatSpecificTransmuxingData (Metadata.Entry, media3-container) carries a raw box payload for transmuxing.
• Extractor attaches the dec3 payload to Format.metadata (not initializationData), so the E-AC-3 decode path is unchanged.
• Muxer writes the dec3 box / ec-3 fourcc from that metadata entry; AUDIO_E_AC3 / AUDIO_E_AC3_JOC added to Mp4Muxer + FragmentedMp4Muxer supported types, so Transformer can transmux Atmos.
• Tests: BoxesTest + end-to-end transmux (Mp4MuxerEndToEndTest, FragmentedMp4MuxerEndToEndTest); extractor golden dumps regenerated.

[FongMi]

Findings

1. P1 Mp4Muxer.java:417 / FragmentedMp4Muxer.java:184
   Adding AUDIO_E_AC3 / AUDIO_E_AC3_JOC to the supported list causes the default Transformer to directly transmux E-AC-3. However, dec3Box() requires the raw dec3 payload in Format.initializationData, while the existing MP4 extractor parses dec3 into Format fields without preserving initialization data. For common E-AC-3/Atmos MP4 or TS inputs, this may change behavior from transcoding to AAC into direct copy, then fail later while muxing moov because csd-0 is missing. I’d recommend either preserving raw dec3 in BoxParser in the same PR, or not adding these MIME types to the factory supported list yet.

2. P3 Boxes.java:1658
dec3Box() only checks that initializationData is non-empty, but does not check that csd-0 itself has bytes. An empty payload would produce an empty dec3 box, which is not a valid EC3SpecificBox. This should mirror esdsBox() / hvcCBox() with checkArgument(csd0.length > 0, ...), plus a test for an empty byte array.

[rexjin-meta]

Hey @FongMi and @snow2405 , Thanks for the reviews — addressed both comments:

• P1 (Transformer regression): Dropped AUDIO_E_AC3 / AUDIO_E_AC3_JOC from SUPPORTED_AUDIO_SAMPLE_MIME_TYPES in both Mp4Muxer and FragmentedMp4Muxer, so the default Transformer keeps transcoding E-AC-3 to AAC — no behavior change. The muxer retains the dec3/ec-3 write capability for direct callers that supply csd-0. Added a comment on dec3Box() explaining the deferral.

• P3 (empty csd-0): Added checkArgument(csd0.length > 0, "csd-0 is empty for dec3 box."), mirroring esdsBox() / hvcCBox(), plus a test for the empty-byte[] case.

Also added a unit test asserting the ec-3 fourcc on the audio sample entry.

This PR will laser focus on enabling muxer to write atmos.
Follow-up (separate PR): preserve the raw dec3 payload in BoxParser, then add E-AC-3 to the supported list and add an end-to-end mux test (with the audio_sample_entry_box_eac3 golden dump). This staging keeps the current PR free of the extractor change and any Transformer impact.

[rexjin-meta]

cc: @andrewlewis This PR is the new attempt for my previously closed PR: #3312 (comment)

[rexjin-meta]

Updated the PR to fix the extractor and muxer together, as suggested.

Extractor: BoxParser now preserves the raw dec3 (EC3SpecificBox) payload in Format.initializationData when parsing the MP4 dec3 box — mirroring the existing av1C/dOps handling (box body without header). This is the piece that was missing.

Muxer: implements the dec3 box + ec-3 fourcc and adds AUDIO_E_AC3 / AUDIO_E_AC3_JOC to the supported audio MIME types. With the extractor change in place, the default Transformer now correctly stream-copies E-AC-3 / Atmos (instead of transcoding), and it no longer risks the missing-csd-0 failure — the two changes are consistent across extractor, muxer, and transformer.

Tests: added an end-to-end Mp4MuxerEndToEndParameterizedTest pass for sample_eac3.mp4 and sample_eac3joc.mp4; regenerated the extractor golden dumps (they now include the dec3 initializationData); BoxesTest covers dec3 wrapping, the ec-3 fourcc, and the missing/empty csd-0 guards.

[rexjin-meta]

Reworked per your suggestion. Added FormatSpecificTransmuxingData (a Metadata.Entry in media3-container, boxType + raw data). The extractor now attaches the dec3 payload to Format.metadata instead of initializationData, so the E-AC-3 decode path is unchanged; the muxer reads it back from Format.metadata to write the dec3 box.

Tests: BoxesTest covers dec3 wrapping, the ec-3 fourcc, and missing/empty-payload guards; Mp4MuxerEndToEndTest + FragmentedMp4MuxerEndToEndTest do a full Mp4Extractor → mux → re-extract round trip for mple_eac3(.mp4|joc.mp4). Extractor golden dumps regenerated.

Made the entry general (boxType + data) rather than E-AC-3-specific — open to either.

[snow2405]

Hey @rexjin-meta
I'm going to send this for internal review now. You may see some more commits being added as I make changes in response to review feedback. Please refrain from pushing any more substantive changes as it will complicate the internal review - thanks!