fix: Replace flaky performance tests with statistical benchmarking fr…

[braedonsaunders]

…amework

Introduces a world-class performance testing approach that eliminates test flakiness caused by JIT compilation, system load, and CI environment variance.

Key changes:

• Add BenchmarkRunner.ts: Statistical benchmark harness with warmup phases, outlier removal (IQR method), and percentile-based metrics (p50/p75/p95)
• Add performanceTestHelpers.ts: Utilities for algorithmic complexity testing, cache effectiveness validation, and adaptive threshold management
• Refactor flockingBehavior.perf.test.ts: Replace single performance.now() measurements with statistical sampling and environment-calibrated thresholds
• Refactor gameLoop.test.ts: Use Vitest fake timers for deterministic timing tests instead of real setTimeout
• Add @tests alias to vitest.config.ts for test utility imports
• Document performance testing guidelines in docs/TESTING.md

The new approach:

• Multiple iterations with warmup phases (JIT optimization)
• Statistical metrics (median, p95) instead of single measurements
• Environment calibration for CI vs local differences
• Algorithmic complexity verification (O(n) vs O(n²)) instead of absolute times
• Welch's t-test for cache effectiveness validation

https://claude.ai/code/session_01PPES1DBUDZVb8fzcoe9Rwh