feat(moq-mux): carry B-frame reorder depth as catalog jitter (exact DTS reserve)

[kixelated]

Stacked on #1843. Makes the exporter's decode-clock reserve exact instead of a fixed guess, so DTS ≤ PTS holds for real B-frame feeds — by carrying the reorder depth the source already knows through the catalog.

What

• TS import now reads the PES DTS (previously discarded), computes the reorder delay (max(PTS − DTS)), and folds it into the video rendition's catalog jitter. The jitter field already documents this exact case ("up to 3 b-frames in a row → 3 × 1000/fps") — this finally populates it. The jitter estimator now reports max(min frame duration, max reorder), and a freshly created rendition is seeded from it so a dirty start (frames before the first SPS, as in the kyrion capture) doesn't lose the value.
• TS export sizes its decode-clock reserve from the catalog jitter, falling back to the small monotonic-only nudge (fix(moq-mux): author DTS for B-frame MPEG-TS export #1843) when none is declared. With the reorder depth known, the authored DTS is both strictly increasing and ≤ PTS.

This rides through the relay via the catalog, so it survives pub | relay | sub with no per-frame wire change, and no catalog format change (jitter already existed — only its population and consumption change).

Why this over the fixed reserve

#1843 keeps DTS monotonic (fixes +igndts) but can't guarantee DTS ≤ PTS without either over-buffering (large fixed reserve = latency) or guessing. The source TS states the exact reorder (PTS − DTS); this carries it, so the reserve equals the actual reorder depth — minimal latency, exact, and B-frame-free streams are unaffected (reserve ≈ one frame).

Notes / decisions for the reviewer

• Final branch: I based this on the fix(moq-mux): author DTS for B-frame MPEG-TS export #1843 branch (not dev) only to keep the diff to one commit — dev is currently ~35 commits behind main, so a direct dev PR would show all of that. It's not a catalog format change, so it can land on main together with fix(moq-mux): author DTS for B-frame MPEG-TS export #1843, or be retargeted to dev if you'd rather batch it there. Your call at merge time.
• Behavior change: B-frame streams now report a larger jitter (the reorder delay, e.g. ~167 ms for kyrion) instead of just the frame interval — which is the field's documented meaning, and correctly sizes the browser player's buffer too.
• Re-import unaffected: the importer still delivers frames in decode order with PTS; the DTS is only used to derive the reserve.

Test plan

• author_dts: monotonic across reorder with the fallback; DTS ≤ PTS once the reserve covers the reorder span.
• Import: import_kyrion_video_jitter_captures_reorder asserts the video jitter reflects the ~5-frame reorder (~167 ms), not the ~33 ms frame interval.
• End-to-end: the kyrion export test now asserts the video PES carry a strictly increasing DTS ≤ PTS timeline (the whole import → jitter → export chain).
• Full moq-mux suite (249 tests), cargo clippy, cargo fmt clean (via nix).

(Written by Claude)