Add comprehensive GC Minimization Plan for performance recovery

Problem: 9.3% performance regression (4.62s vs 4.20s baseline) after GC modularization
Root Cause: Module boundaries introduced function call overhead, lost compiler optimizations

This plan provides 6-phase strategy to recover performance:
- Phase 0: Profiling (25-30h) - Identify actual hotspots
- Phase 1: Hot Path Inlining (20-25h) - Expected 3-5% recovery
- Phase 2: Compiler Optimizations (5-8h) - Expected 1-2% recovery
- Phase 3: Allocation Fast Path (15-20h) - Expected 1-2% improvement
- Phase 4: GC Parameter Tuning (15-20h) - Expected 1-2% improvement
- Phase 5: Advanced Research (40-50h) - Experimental

Target: ≤4.33s (3% tolerance)
Conservative estimate: 4.26s with Phases 0-2 (50-63 hours)
Optimistic estimate: 4.18s with Phases 0-3 (65-83 hours)

Key insights:
- Profile first, optimize second (don't guess)
- Inline top 10 hot functions identified by profiling
- Enable LTO for cross-module optimization
- Batch GC checks to reduce allocation overhead
- Tune GC parameters to reduce frequency

Next steps: Begin Phase 0 (Profiling) immediately

Author: claude