Add prompt caching support for AWS Bedrock Converse API

[sobychacko]

Implements prompt caching to reduce costs on repeated content and improve
response times. Applications with large system prompts, extensive tool
definitions, or multi-turn conversations can see significant savings, as
cached content costs ~90% less to process than uncached content.

Adds five caching strategies to address different use cases:

• SYSTEM_ONLY: Cache system messages (most common - stable instructions)
• TOOLS_ONLY: Cache tool definitions (when tools are stable but system varies)
• SYSTEM_AND_TOOLS: Cache both (when both are large and stable)
• CONVERSATION_HISTORY: Cache conversation history (for chatbots and assistants)
• NONE: Default, no caching

Implementation:

• BedrockCacheStrategy enum with BedrockCacheOptions configuration class
• Integrated with BedrockChatOptions (equals/hashCode/copy support)
• Cache points applied as separate blocks to satisfy AWS SDK UNION type constraints where each block can only contain one field type
• Boolean flags derived from strategy to improve code readability and avoid repetitive conditional checks throughout request building
• Last user message pattern for CONVERSATION_HISTORY enables incremental caching where each turn builds on the previous cached prefix
• Cache metrics exposed via metadata Map to maintain provider independence without adding Bedrock-specific fields to shared interfaces
• Cache hierarchy respects AWS cascade invalidation (tools → system → messages) to prevent stale cache combinations
• Debug logging for troubleshooting cache point application

Model compatibility:

• Claude 3.x/4.x: All strategies supported
• Amazon Nova: SYSTEM_ONLY and CONVERSATION_HISTORY only (AWS limitation on tool caching for Nova models)

Testing:

• Integration tests for all strategies using Claude 3.7 Sonnet
• Tests handle cache TTL overlap between runs to avoid flakiness in CI

Documentation includes usage examples, real-world use cases (legal document
analysis, code review, customer support, multi-tenant SaaS), best practices,
cache invalidation behavior, and cost considerations. Break-even occurs after
one cache hit since cache reads cost ~90% less than base input tokens while
cache writes cost ~25% more.

Thank you for taking time to contribute this pull request!
You might have already read the contributor guide, but as a reminder, please make sure to:

• Add a Signed-off-by line to each commit (git commit -s) per the DCO
• Rebase your changes on the latest main branch and squash your commits
• Add/Update unit tests as needed
• Run a build and make sure all tests pass prior to submission

For more details, please check the contributor guide.
Thank you upfront!