Audit round 1: playback order, silent title drops, image durability

Four fixes from the first audit round. Every finding was verified against a
pinned tree and read directly before being accepted.

resolve_vts_key sorted a VTS title-VOB extents largest-first. That is the 1.5.1
garbage bug, and it grew back in a new code path: the comment claimed it
"matched the scan heuristic", but that heuristic WAS the bug and had already
been fixed in decrypt_keys_for_title, which documents the rule (PLAYBACK ORDER,
never largest-cell-first) and pins it with a regression test. A CSS DVDs biggest
cell opens with a long clear run and crack_key shares one sector budget across
the extent list, so starting there can exhaust it without ever MEETING
scrambled data — and CSS recovers the key from scrambled data itself. The crack
then returns None, the caller falls back to the disc-wide key, and every VOB in
that VTS is descrambled wrongly: corrupt PES behind an intact header, written
out as a complete extract at exit 0.

parse_pgcit dropped titles silently in THREE places — an unparseable PGC, an
out-of-range PGC index, and a truncated entry table. The finder caught one; the
other two turned up on reading the function. parse_vmg already counts and warns
per skipped title SET for exactly this reason, and this was the last place a
disc could quietly report fewer titles than it has.

write_image called flush() and returned Ok. flush() only pushes bytes into the
page cache and promises nothing about durability, so a 6-90 GB image could be
reported complete while still unwritten — a crash or an unmounted volume then
leaves a truncated file the caller was told was finished. Now into_inner (so a
buffered-write error surfaces instead of being dropped by BufWriter::drop)
followed by sync_all.

timeline used abs() on a saturating_sub result. Every other comparison in that
module is saturating because the timestamps come off a disc and are not
trusted; abs() panics on i64::MIN, which saturating_sub can produce.

src/mux/demux_sink.rs

@@ -190,7 +190,13 @@ pub(crate) fn codec_label(codec: Codec) -> &'static str {
 /// to `.sup`, VobSub records `.idx` entries and emits the sidecar at finish.
 trait EsWriter: Send {
     /// Write one frame's payload to `w`. Returns the number of bytes written to
-    /// the main file (used by the VobSub writer for `.idx` filepos tracking).
+    /// the main file.
+    ///
+    /// Nothing in production reads this count: the sole caller discards it with
+    /// `?`, and `VobSubWriter` tracks `.idx` fileposes from its own `pos` field.
+    /// This previously claimed the VobSub writer depended on the return value,
+    /// which would have a maintainer believe a wrong count shows up as broken
+    /// `.idx` output. It does not.
     fn write_frame(&mut self, w: &mut dyn Write, f: &PesFrame, pts_ns: i64) -> io::Result<usize>;
 
     /// Finalize. Default: no-op. The VobSub writer serializes its `.idx` here.

src/mux/mkv.rs

@@ -680,8 +680,9 @@ pub struct MkvMuxer<W: Write + Seek> {
     cues: Vec<CuePoint>,
     frame_count: u64,
     /// Frames handed to `write_frame` that were dropped because no cluster was
-    /// open yet (a cluster only opens on a track-0 video keyframe). See
-    /// `write_frame` for the track-0 invariant.
+    /// open yet (a cluster only opens on a keyframe from the primary video
+    /// track, whatever index that is — not necessarily track 0). See
+    /// `write_frame` for the cluster-driver invariant.
     ///
     /// The ALL-dropped case is surfaced as an error by `finish()`, but via
     /// `frame_count == 0`, not via this counter. A PARTIAL drop — leading audio /
@@ -1683,11 +1684,17 @@ impl<W: Write + Seek> MkvMuxer<W> {
 
     /// Finish the MKV file: write Cues element.
     ///
-    /// # Track-0 invariant
+    /// # Cluster-driver invariant
     ///
-    /// A cluster only opens on a track-0 video keyframe, so the caller must
-    /// supply track 0 as the video track and deliver a keyframe on it before
-    /// (or alongside) other-track data. If no track-0 keyframe ever arrives,
+    /// A cluster only opens on a keyframe from the PRIMARY VIDEO TRACK — the
+    /// first track whose type is video, at whatever index it occupies (see
+    /// `cluster_driver`, which is `primary_video_track.unwrap_or(0)`; index 0
+    /// is only the fallback for a file with no video track at all). The caller
+    /// must deliver a keyframe on that track before (or alongside) other-track
+    /// data. This said "track-0" and required the caller to place video at
+    /// index 0, which the code has never actually required — so a reader
+    /// debugging dropped frames would suspect track ordering, which is not it.
+    /// If no such keyframe ever arrives,
     /// every `write_frame` is silently dropped; rather than emit a structurally
     /// valid but empty MKV (zero clusters, zero frames), `finish` returns
     /// `Error::MkvInvalid` when frames were submitted but none were written.

src/mux/timeline.rs

@@ -264,7 +264,13 @@ impl SeamPlan {
             // the end of the list.
             let stepped_back = pos.last_raw_ns.is_some_and(|last| raw_ns < last)
                 && if has_reorder {
-                    (raw_ns.saturating_sub(next_in)).abs() <= CLIP_START_TOLERANCE_NS
+                    // saturating_abs, not abs: every other comparison in this
+                    // module is saturating because these timestamps come off a
+                    // disc and are not trusted. `abs()` is the one exception
+                    // and it panics on i64::MIN — which saturating_sub can
+                    // produce exactly — taking down the mux thread on one bad
+                    // frame instead of comparing false like its neighbours.
+                    (raw_ns.saturating_sub(next_in)).saturating_abs() <= CLIP_START_TOLERANCE_NS
                 } else {
                     raw_ns >= next_in.saturating_sub(CLIP_START_TOLERANCE_NS)
                         && raw_ns <= self.clips[clip + 1].out_ns