[Caideyipi]

Summary

Add bounded, low-overhead diagnostics for subscription replay stalls and WAL footer recovery after abrupt termination.

What changed

• Reuse the existing stalled-replay warning throttle and record the actual writer barrier that blocks progress.
• Include bounded admission/memory, buffered-head, search-index, writer-progress, safe-frontier, retention, and WAL-iterator state in the stall snapshot.
• Sample at most four writers per diagnostic group, prioritizing unknown and blocking writers.
• Emit one bounded WARN only after a WAL footer is actually repaired, summarizing file size, entry count, writer progress, and a few samples.
• Keep normal write/replay behavior unchanged: no per-entry logging, WAL format changes, extra fsync, or retention/ACK changes.

Validation

• English full-reactor test-compile -DskipTests
• Targeted WAL repair, prefetch queue, writer ordering, and progress iterator tests
• Spotless formatting and git diff --check

The original production root cause still needs confirmation from the next reproduction using these diagnostics.