Stress: Concurrent read+write benchmark on large indexed table — measure p99 degradation ceiling

Evidence: large outdoor retail redirect cluster redirect lookup (multi-condition scan on 1.44M-record 5-index table) runs under concurrent bulk CSV import. Add benchmark: concurrent readers do multi-condition index scan on 1M-record 3-index table while writers insert at 1K rows/sec; measure read p99 during write load vs. at rest; assert degradation below ceiling. Confluence: Section 5 ST-5

## Existing coverage in harper-pro
The `integrationTests/stress/` suite in harper-pro covers related concurrent-read-under-write scenarios:
- `slowConsumerBackpressure.test.mjs` — asymmetric peer (1 thread vs 4) under write churn
- `soakWithRollingRestarts.test.mjs` — sustained writes with concurrent node restarts

This issue should focus on the indexed-table-specific scenario (multi-condition scan on a large table with 5 secondary indexes under bulk import load) which is not covered by the replication-focused stress suite.