Measure and optimize token exchange performance

[jhrozek]

Profile token exchange performance in Virtual MCP Server and implement token caching if measurements show significant latency impact.

This is a deferred optimization from epic #150 (in epics repo).

Scope

• Benchmark token exchange latency in realistic scenarios
• Implement token cache if measurements justify the added complexity
• Cache invalidation based on token expiry
• Thread-safe cache implementation

Acceptance Criteria

• Performance measurements documented (latency, throughput)
• Decision documented: cache needed or not needed with rationale
• If cache implemented: reduces token exchange calls significantly
• If cache implemented: respects token expiry
• If cache implemented: thread-safe under concurrent load