Implement request caching and rate limiting

[cablate]

Overview

Every MCP tool call currently makes a live Google Maps API request. Adding a cache layer and rate limiting will reduce latency, lower API costs, and prevent quota exhaustion.

Proposed Work

Caching

• Introduce an in-memory LRU cache (e.g., lru-cache) keyed on the serialized request parameters
• Configurable TTL per tool (e.g., place details can be cached longer than nearby search)
• Cache bypass via an optional noCache parameter or environment variable

Rate Limiting

• Respect Google Maps API quotas (requests per second / per day)
• Implement a simple token-bucket or sliding-window rate limiter
• Return a graceful error message when the rate limit is reached rather than crashing

Configuration

• Expose cache TTL and rate limit settings via environment variables with sensible defaults

Acceptance Criteria

• Repeated identical requests within TTL are served from cache without a network call
• Exceeding the configured rate limit returns a structured error to the MCP client
• All settings have documented defaults in README