Implement token store and caching for exchanged tokens

Add support in ToolHive proxy for caching and storing exchanged downstream tokens. The goal is to avoid re-exchanging tokens on every call while ensuring tokens are rotated before expiry and never stale.

Key requirements:

• Support two modes: Per-call mode (no caching, always perform a token exchange), Session caching mode (reuse a token within a session, keyed by {subject_token_hash, audience, scopes}).
• Implement configurable TTL and proactive rotation (e.g. refresh when 80% of token lifetime has elapsed).
• Support multiple storage backends: In-memory (default, single-node), Redis (pluggable option for multi-node deployments).
• Ensure tokens are stored ephemerally only, no persistence beyond configured TTL.
• Add safeguards: Clear tokens on error or explicit expiry, Redact tokens from logs/metrics.

Tasks:

• Implement code to perform this caching of tokens
• [ ] Implement different caching backends (local, redis)
• Unit tests cover cache hit, miss, expiry, and rotation.
• [ ] Integration tests for different methods and storage backends
• [ ] Documentation updates explaining caching strategies, trade-offs, and secure storage practices.

Start by locating the ToolHive proxy path that exchanges downstream tokens and identify how proxy sessions and configuration are represented. Define the boundaries for per-call versus session caching, in-memory versus Redis storage, TTL rotation, and token redaction before implementing. Add unit coverage for cache hits, misses, expiry, and rotation, integration coverage for methods and backends, and document the caching and secure-storage trade-offs.