feat: Production-grade provider manager with resilience patterns

Enhanced provider manager with:

- Circuit breaker pattern: Automatic failover when providers fail
  (threshold: 5 failures, 1min recovery)
- LRU caching: Response caching with TTL and size limits (1000 max)
- Rate limiting: Per-provider token bucket (60/min cloud, 1000/min local)
- Health checks: Periodic provider availability monitoring
- Latency percentiles: p50/p95/p99 tracking for SLA monitoring
- Model-specific pricing: Accurate cost tracking per model variant
- Token extraction: Real usage tracking from API responses

Tests expanded from 10 to 40+ covering all new features.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: SolariSystems