Reuse a full-subset cache hit for pyflow play (TASK-050)

[AdamClemens]

Summary

Replaces #85, which GitHub auto-closed (and then refused to reopen)
when its base branch (feat/checkpoint-retention-policy) was deleted
after #84 merged. Same commit (a746e83), same content, now targeting
main directly since #83/#84 are already merged.

Closes Stage 8's Completion Criterion 9: a pyflow play --cache DIR
request that falls fully inside an already-cached wider window is
sliced from it directly, at zero re-simulation cost.

• _find_superset_window (src/pyflow/replay.py) ranks candidates by
  the range in the filename first, the same cheap-before-I/O shape
  find_checkpoint_at_or_before already uses, then slices the frames
  needed straight out of the matching window.
• No new cache file is written for the sliced sub-range -- only an
  exact-range request still writes one, unchanged from before this.
• A request that only partially overlaps a cached range, or extends
  past its edge, still falls back to full materialize_window -- a
  real, stated exclusion (full stitching across a cached window's own
  edge is deferred future work), confirmed by a deliberate mutation
  observed to fail the regression test before being reverted.

Test plan

• Strict TDD: every new test written first, confirmed red, then
  green
• Mutation test on the partial-overlap exclusion, confirmed to fail
  before being reverted
• make lint and every relevant make check-* target clean
• make test -- 1164 passed, 21 skipped, exit 0
• Verified by hand against the real CLI: caching pyflow play --from-frame 0 --to-frame 20 --cache DIR, deleting every
  checkpoint, then pyflow play --from-frame 5 --to-frame 10 --cache DIR still exits 0 with no new cache file written for the
  narrower range
• Already passed CI green on both platforms as Reuse a full-subset cache hit for pyflow play (TASK-050) #85 (unchanged
  commit)

🤖 Generated with Claude Code