refactor(moq-mux)!: split the fMP4 init segment out of Fragment

[zshenker]

Summary

fmp4::Fragment described two different emits with one struct. The init segment carried init: true, independent: false, and duration: 0.0; every media fragment carried init: false and meant the other two. Three of its four fields were sentinels on one of the two shapes it covered, and a consumer had to read a bool to know which fields were real.

Export::next_chunk now returns a Chunk: Init holds the bytes and nothing else, while Fragment holds a Fragment with no dead field. Export::next is unchanged for consumers that only want bytes.

Review feedback also moved Fragment::duration from untyped f64 seconds to std::time::Duration, and added consuming Chunk::init and Chunk::fragment accessors. rs/CLAUDE.md now records the repository convention to use typed time values and convert numeric serialized forms at the boundary, using serde_with where possible.

Chunk is deliberately not #[non_exhaustive]. The set is closed by the container format, which emits an init segment and media segments and nothing else, so a wildcard arm would be permanent noise on every match.

Public API changes

item	change
fmp4::Chunk	new enum: Init(Bytes), Fragment(Fragment)
Chunk::data, Chunk::into_data	new byte accessors
Chunk::init, Chunk::fragment	new consuming variant accessors
Fragment::init	removed
Fragment::duration	changed from f64 seconds to std::time::Duration
Export::next_fragment	renamed to next_chunk, now returns Option<Chunk>
Export::poll_next_fragment	renamed to poll_next_chunk, now returns Option<Chunk>
Export::next / poll_next	unchanged

This targets dev because it breaks published moq-mux APIs.

Test plan

Run on the final head after merging the latest origin/dev:

• nix develop --command just fix
• nix develop --command just check
• nix develop --command just rs test -p moq-mux: 482 passed, 0 failed

just test smoke-full was not run because this changes no wire format, moq-ffi API, or gateway behavior.

Cross-Package Sync

No table rows apply. moq-mux has no cross-package row, no wire format or CLI surface changed, and doc/lib/rs/crate/moq-mux.md describes the crate's scope rather than this API.

(Written by GPT-5)

[kixelated]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep it up!

Reviewed commit: 4046b6d3be

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[kixelated]

Reviewed the final head after syncing with dev and adding the typed-time guidance. The chunk API cleanly separates init data from media fragment metadata, the Duration change preserves exact internal values, and the two prior review requests are implemented. just check and all 482 moq-mux nextest tests passed. No blocking findings.

(Written by GPT-5)

[kixelated]

Reviewed the final head after merging the latest dev. The upstream commit is unrelated to the fMP4 API change and merged without conflicts. The full Rust CI test stage passed all 2,955 tests locally, the scoped repository checks passed on the prior equivalent head, and the final diff passes git diff --check. No blocking findings.\n\n(Written by GPT-5)