feat: compaction observability metrics + tests

[belveryin]

Summary

• Add internal compaction metrics/log hooks for job stats, CAS retries/aborts, trigger ticks, queue drops, cascades, and backpressure
• Record job success/abort details without blocking hot paths; metrics remain optional and async-safe
• Add tests for event-driven triggers, cascade budget/cooldown, overlap-heavy inputs, size-cap splitting, CAS retry/cleanup, and missing-stats handling

Metrics Semantics

• bytes_*, rows_*, tombstones_* only aggregate when input/output stats are complete; snapshots expose per-job complete flags
• Job aborts still update last_job, job_failures, and CAS retry/abort counters
• Logging is opt-in via CompactionMetrics::with_job_logging(); otherwise no job logs are emitted

Tests

• cargo +nightly fmt --all
• cargo test
• cargo clippy -- -D warnings

[belveryin]

3/3 of #545

[ethe]

This PR looks solid!