Disables the seq_bypass optimization which harmed real-world workloads.
Benchmarks (4GB SQLite, full table scan, cold cache, interleaved OFF/ON):
- OFF avg 52.58s → ON avg 43.12s → 18.0% speedup
- 3 passes, module reloaded between each OFF/ON pair
- hit_rate_real: 99%, stride_predictions: 5.74M per pass
Changes:
- Commented out seq_bypass return statements in nsd.c (stride-bypass block, workload update, hot update)
- Code comments translated to English
- SQLite full-table-scan benchmark script added