Add distributed caching for GCP ID tokens

[reakaleek]

Problem

Lambda containers were each generating GCP ID tokens independently.
When many containers start at once, they all call Google OAuth at the same time,
causing rate limit errors.

Solution

Implemented two-tier distributed cache to share tokens across all Lambda containers:

• L1 (in-memory): Sub-millisecond access for warm Lambda invocations
• L2 (DynamoDB): Cross-container sharing - first container generates token,
  others reuse it (prevents all containers calling Google at once)

Changes

• New cache system: IDistributedCache, MultiLayerCache, DynamoDbDistributedCache
• Updated GcpIdTokenProvider to use cache instead of static dictionary
• Environment-aware: In-memory for dev, DynamoDB for prod/staging/edge
• Token reused for 45 minutes (15-min safety buffer)
• Added OTel tracing and structured logging

Infrastructure Required

DynamoDB table: docs-api-cache-${environment} with CacheKey partition key and TTL enabled.
Lambda IAM: dynamodb:GetItem + dynamodb:PutItem permissions.