feat(mp4): add the rsot box for redundant sample original timing

[tobbee]

Adds the Redundant Sample Original Timing Box (rsot), new in ISO/IEC 14496-12:2026
§8.8.18, as a traf child.

What it is

Box Type:  'rsot'
Container: TrackFragmentBox
Quantity:  Zero or one

aligned(8) class RedundantSampleOriginalTimingBox extends FullBox('rsot', 0, flags) {
   if (flags & rsot_original_duration)  unsigned int(32) original_duration;
   if (flags & rsot_elapsed_duration)   unsigned int(32) elapsed_duration;
}

It documents two things about the edges of a track fragment:

• elapsed_duration (flag 0x2) — the first sample of the fragment is a copy of the
  previous sample, and has already been presented for this long. A receiver that already
  has the previous sample extends that sample by this one's duration and ignores
  elapsed_duration; one that tunes in here instead presents it as if it had already run
  for elapsed_duration. When this flag is set, that first sample shall carry
  sample_depends_on = 2 and sample_has_redundancy = 1.
• original_duration (flag 0x1) — the last sample of the fragment was truncated to
  respect fragmentation constraints, and was originally authored this long.

The spec's NOTE 1 names the case it exists for: "In adaptive streaming context where media
segments of fixed duration are not aligned with variable frame rate media such as text,
possible duplicated redundant samples may happen at segment boundaries." That is the
subtitle-cue-across-a-chunk-boundary case — a wvtt cue or stpp document restated in the
next fragment, where the container previously had no way to say "this is a continuation,
and it really runs until T".

API

rsot := mp4.CreateRsotBox(originalDuration, elapsedDuration) // 0 means "not signalled"
rsot.HasOriginalDuration()
rsot.HasElapsedDuration()
traf.AddChild(rsot)
traf.Rsot // named link, like Tfdt and Senc

Info renders:

[rsot] size=20 version=0 flags=000003
 - originalDuration: 2000
 - elapsedDuration: 500

Two decisions worth a look

1. CreateRsotBox derives the flags from its non-zero arguments rather than taking a
   separate flags argument. §8.8.18.3 says a signalled duration "shall be different from
   0", so zero cannot mean anything else, and this makes an inconsistent flags/field pair
   unconstructible. Both zero gives a legal 12-byte box with no flags set.

2. Decode rejects a box whose size does not match its flags. Since both fields are
   optional and independent, the size follows exactly from the flags once they are read, so
   any other size means a malformed box. Without the check a truncated rsot decoded
   silently with a zeroed duration. This follows the clli convention
   (invalid box size %d) rather than tfhd, which does no size check.

Files

File	Change
mp4/rsot.go	new RsotBox with the full Box interface, DecodeRsot/DecodeRsotSR, CreateRsotBox, HasOriginalDuration, HasElapsedDuration
mp4/box.go, mp4/boxsr.go	"rsot" in decoders and decodersSR
mp4/traf.go	TrafBox.Rsot plus the AddChild case
mp4/rsot_test.go	all four flag combinations through boxDiffAfterEncodeAndDecode, byte-exact hex goldens, changeBoxSizeAndAssertError, and a traf round trip checking the named link
CHANGELOG.md	Unreleased / Added

Follows the "How to implement a new box" guide in the README. go test ./..., gofmt and
golangci-lint are clean.

🤖 Generated with Claude Code