audit: bound DTS marks, align disc-format tree order, doc/test cleanups

Round-7 findings from the 10-phase release audit (no HIGH; convergence):

- Cap DtsParser.pts_marks (MAX_PTS_MARKS): a run of zero-length timed PES
  packets grew no buffer bytes, so the drain_front mark-prune never ran —
  the deque could accumulate without bound on hostile PS input.
- detect_disc_format tested HVDVD_TS before BDMV while the title-scan
  dispatch tests BDMV first, so a disc with both trees would be classified
  HD-DVD but enumerated as Blu-ray. Align both to BDMV → HVDVD_TS →
  VIDEO_TS.
- Document why the DTS new-PES re-base can emit a locally-decreasing PTS
  (the muxer's block_ts applies the strictly-monotonic audio nudge, tested
  in mkv.rs) — this is by design, not a mux defect.
- Fix stale aacs/keys.rs comment references (functions moved to
  aacs/inf.rs / aacs::resolve/derive in the module split).

src/aacs/variant.rs

@@ -889,7 +889,7 @@ mod tests {
     fn mkb_records_matches_walk_mkb_framing() {
         // The lazy `mkb_records` iterator and the owning `walk_mkb` must agree on
         // (offset, type, len) for every record — they share the one framing
-        // walker, and every keys.rs MKB walk now relies on this equivalence.
+        // walker, and every aacs::resolve/derive MKB walk now relies on this equivalence.
         let mut mkb = vec![0x10, 0x00, 0x00, 0x06, 0xAA, 0xBB];
         mkb.extend_from_slice(&[0x05, 0x00, 0x00, 0x08, 1, 2, 3, 4]);
         mkb.extend_from_slice(&[0x00, 0x00, 0x00, 0x00, 0xFF]); // terminator + trailing