Flagged-bail fixture for the match_mark queue-dispatch park.
See docs/heuristics_audit.md (the two narrow defensive bails still flagged).

WHAT IT EXERCISES
  render/cascade.rs, in drive_leaf's QueueDispatch arm, the
  `if self.match_mark.is_none()` guard that records the round-boundary cut
  for a cross-leaf queue dispatch. This fixture deterministically reaches
  that site: `provider` emits an out-event callback (`p.cb`) to its
  suspended cross-leaf caller `client`; `client` parks the callback and
  dispatches it as its own QueueDispatch hop, which sets the match_mark.

WHY simulate.0.override (sanctioned ours-as-golden), not XFAIL
  nucleos renders the round fully and reaches the same RTC boundary as dzn;
  the ONLY divergence is the requires-port mirror value (`client.p.busy`) on
  two INTRA-RTC `--strict` state lines inside the parked-dispatch round —
  nucleos commits the mirror eagerly, dzn lags it. RTC-boundary states agree
  (dzn graph -b state). Per the standing ruling [[queue_lag_vs_display_bug]],
  a sync-only stale requires-mirror is a dzn `--strict` checkpoint DISPLAY bug
  (nucleos is canonical) and must NOT be fitted. dzn graph -b state cannot
  arbitrate this SYSTEM (dzn 2.20.0 collapses it to a single node), so
  canonicality rests on RTC-boundary agreement + the never-fit-stale-mirror
  rule. User-authorized per-model override (2026-06-28). simulate.0.override
  holds nucleos's canonical output; simulate-repl shares it (repl == batch).

lts.skip
  dzn 2.20.0 in this toolchain does not emit a composed SYSTEM LTS via
  `verify --out=aut` (silent empty output; works for leaf components only),
  so no dzn lts.aut baseline could be generated here. verify-only fixture;
  `nucleos -v verify -k` is byte-identical to dzn (baseline/verify).
