Feature/agent cache adding OpenAI anthropic llamaindex adapters

[KIvanow]

Summary

Changes

Checklist

• Unit / integration tests added
• Docs added / updated
• Roborev review passed — run roborev review --branch or /roborev-review-branch in Claude Code (internal)
• Competitive analysis done / discussed (internal)
• Blog post about it discussed (internal)

---

Note

Medium Risk
Adds new provider adapters and changes LLM cache key computation/entry shape to support multimodal content blocks, which could affect cache hit rates and compatibility for existing consumers.

Overview
Adds multi-provider, multimodal LLM caching to @betterdb/agent-cache (v0.3.0). Introduces new adapters for openai (Chat Completions), openai-responses, anthropic (Messages), and llamaindex, each normalizing provider-specific inputs into a shared canonical LlmCacheParams format.

Extends cache key + storage model for richer responses. llmCacheHash now includes additional request fields (toolChoice, seed, stop, responseFormat, reasoningEffort, promptCacheKey) and new exported ContentBlock types; the LLM tier adds storeMultipart() to persist structured contentBlocks while still flattening text into the legacy response field, and check() can now return contentBlocks when present.

Adds pluggable binary normalization and examples/tests. Introduces BinaryNormalizer utilities (hashBase64, hashUrl, fetchAndHash, composeNormalizer, etc.), adds example apps for OpenAI/Anthropic/LlamaIndex, expands unit tests/snapshots for adapter normalization and cross-provider hash parity, and updates the release workflow to verify the newly built adapter artifacts.

^{Reviewed by Cursor Bugbot for commit 9faff5c. Bugbot is set up for automated code reviews on this repo. Configure here.}