[Critical] Add latency benchmarks and SLA targets

[jessfortemnaturae8717]

Problem

No performance benchmarks or SLA targets exist. We can't promise customers any latency guarantees.

Current state:

• TraceBuilder tracks duration_ms but nothing consumes it
• No vitest benchmarks
• No CI gates for performance regression
• No p95/p99 measurements

Risk

• Can't set SLA for Pro/Enterprise tiers
• Performance regressions go unnoticed
• Customers experience unpredictable latency

Solution

Add comprehensive benchmarking:

// bench/evaluation.bench.ts
import { bench, describe } from 'vitest';

describe('evaluation latency', () => {
  bench('small plan (10 resources)', async () => { ... });
  bench('medium plan (100 resources)', async () => { ... });
  bench('large plan (1000 resources)', async () => { ... });
  bench('with state lookups (mocked)', async () => { ... });
});

Tasks

• Add vitest bench configuration
• Create benchmark suite for evaluation latency
• Benchmark state lookup latency (mocked + real)
• Add CI job to run benchmarks on PR
• Set performance gates (fail if p95 > threshold)
• Add latency histogram to /metrics endpoint
• Document SLA targets in docs

SLA Targets (Proposed)

Scenario	p50	p95	p99
Small plan (≤10 resources)	< 50ms	< 100ms	< 200ms
Medium plan (≤100 resources)	< 200ms	< 500ms	< 1s
Large plan (≤1000 resources)	< 1s	< 3s	< 5s
With live state (per resource)	+50ms	+100ms	+200ms

---

Priority: P1 — Required for Pro tier launch