Bump benchmark iterations from 5 to 30 and wire up workflow input

[Mossaka]

Problem

The benchmark script (scripts/ci/benchmark-performance.ts) hardcodes 5 iterations per metric (line 18). With only 5 samples, P95/P99/max are all the same value — there's no statistical significance in the percentile calculations:

// P95 with n=5: Math.floor(5 * 0.95) = 4, which is just the max value
p95: sorted[Math.min(Math.floor(n * 0.95), n - 1)]

Additionally, the iterations input exists in the workflow dispatch but the script ignores it — it's always 5.

Changes

1. Bump default iterations from 5 to 30 in scripts/ci/benchmark-performance.ts line 18
2. Wire up the workflow dispatch input so iterations is passed to the script (e.g., via env var or CLI arg)
3. Increase workflow timeout from 30 to 60 minutes to accommodate more iterations (30 iterations × ~30s per warm run ≈ 15 minutes for warm alone)

With 30 iterations, P95 = index 28 (the 2nd-worst run), giving meaningful tail-latency data.

Files to Modify

• scripts/ci/benchmark-performance.ts — read iteration count from env var, default to 30

• .github/workflows/performance-monitor.yml — pass input to script, increase timeout

• Epic: [Long-term] Add performance benchmarking suite #240

• Historical storage: [plan] Add historical benchmark storage and relative regression detection #1760

• Daily schedule: [aw] Secret Digger (Copilot) failed #1853

• Startup analysis: Container startup benchmarks fail: 15+ seconds of idle wait in warm startup path #1850

[Mossaka]

Part of benchmark epic #240