Add prompt caching support for BedrockConverse provider

[qrush]

Summary

• Add BedrockConverse::Content helper class for building content with cache points
• Extract cache metrics from responses (cacheReadInputTokenCount, cacheWriteInputTokenCount)
• Preserve Raw content with cachePoint blocks in system messages

Bedrock Converse Caching Format

{
  "system": [
    { "text": "System instructions" },
    { "cachePoint": { "type": "default", "ttl": "5m" } }
  ]
}

Usage

# Cache system instructions
system = RubyLLM::Providers::BedrockConverse::Content.new(
  "You are a helpful assistant",
  ttl: '1h'
)

chat = RubyLLM.chat(model: 'anthropic.claude-3-5-haiku-20241022-v1:0', provider: :bedrock_converse)
chat.with_instructions(system)

response = chat.ask("Hello")
puts response.cached_tokens          # Tokens read from cache
puts response.cache_creation_tokens  # Tokens written to cache

Test plan

• Add specs for BedrockConverse::Content helper
• Update specs for cache token extraction in streaming
• Add specs for cache metrics in sync response parsing
• Add specs for cachePoint preservation in system content

🤖 Generated with Claude Code