libfreemkv: rc.5.2 DVD test coverage — depth-aware mux, colour codes, CSS scan

Implements the rc.5.2 quick-units list from the DVD coverage audit and
corrects the "passes-but-encodes-the-bug" tests that could not
distinguish correct from wrong behaviour.

New tests (each with the bug it guards):

mux/mkv.rs
- field_duration_is_direct_trackentry_child_not_in_video: depth-aware
  check that DefaultDecodedFieldDuration (and DefaultDuration) are direct
  TrackEntry children, NOT nested in the Video master. Replaces the flat
  find_id byte-scan that passed either way. Adds master_children /
  first_track_entry depth-walking helpers.
- pal_576i_emits_bt470bg_colour_codes / ntsc_480i_emits_smpte170m_colour_codes:
  assert the actual CICP tuples written into the MKV Colour master —
  PAL (5,5,5,1) vs NTSC (6,6,6,1) — not just stream-layer ColorSpace.
- ntsc_480i_field_order_is_tff_and_encoded: pins NTSC 480i hardcoded TFF
  and its ~33.37ms/16.68ms frame/field durations, asserting the encoded
  FlagInterlaced/FieldOrder bytes (480i was never exercised before).

ifo.rs
- video_attr_absolute_bytes_pin_real_layout: drives parse_video_attr with
  HARDCODED real DVD-Video bytes (PAL/NTSC x 4:3/16:9, plus mpeg_version
  in bits 7-6) instead of v_atr_byte, so a co-edit of the shift constants
  can't re-seed the PAL-as-NTSC bug. Anchors that permitted_df bits (1-0)
  are not read as the TV system.

disc/dvd.rs
- scan_dvd_titles_lpcm_routes_to_a0_pid_range: LPCM (coding 4) → sub-id
  0xA0 → PID 0xBDA0, disjoint from the AC-3 0xBD8x space, channels kept.
- scan_dvd_titles_multiple_vobsub_tracks_distinct_pids: three VobSub
  tracks → distinct 0x20+ordinal PIDs, per-language, shared palette.

css/mod.rs
- crack_outcome_reaches_cracked_with_span: drives the full crack scan to
  CrackOutcome::Cracked via a Stevenson-crackable synthetic sector and
  asserts crack_span recording (the Cracked branch was never exercised).
- recrack_succeeds_on_other_vts_extents: per-VTS re-crack SUCCESS path.
- all_locked_synthetic_iso_yields_css_key_missing_signal: all-locked
  multi-extent ISO → ScrambledUncracked, the signal the scan converts to
  css_error = Some(CssKeyMissing).

Corrected fixtures (passes-but-encodes-the-bug):
- scan_dvd_titles_mixed_audio_codecs_distinct_pids: real channel nibbles
  (AC-3 5.1 = 6ch, DTS 2.0 = 2ch) replacing the 1ch placeholders; asserts
  channel counts and exact canonical PIDs (0xBD80 / 0xBD88).
- ebml.rs FieldOrder comment: drop the stale "PAL DVD (576i) is
  bottom-field-first" line that contradicted the TFF-for-all code.

src/mux/ebml.rs

@@ -436,8 +436,10 @@ pub const FIELD_ORDER: u32 = 0x9D;
 pub const INTERLACED_INTERLACED: u64 = 1;
 pub const INTERLACED_PROGRESSIVE: u64 = 2;
 // FieldOrder values (Matroska): 0/2 = top-field-first, 1/9 = bottom-field-first.
-// NTSC DVD (480i) and HD (1080i) are top-field-first; PAL DVD (576i) is
-// bottom-field-first. 0xFF is our sentinel for "undetermined / omit".
+// NTSC DVD (480i), PAL DVD (576i) and HD (1080i) are all emitted top-field-first
+// — the muxer hardcodes TFF for every interlaced DVD/HD source (DV is the only
+// common BFF source and freemkv does not produce it). 0xFF is our sentinel for
+// "undetermined / omit".
 pub const FIELD_ORDER_TFF: u8 = 2;
 // Bottom-field-first. Retained for completeness/round-trip tests; the muxer
 // emits TFF for all DVD/HD interlaced content (DV is the only common BFF