[Bifrost] Improve documentation for Weaviate setup

Was calling OpenAI and Anthropic directly from multiple services. Each service had its own API key management, retry logic, and error handling. It was duplicated everywhere and none of it was consistent. Wanted a single proxy that all services call, which handles routing, failover, and rate limiting in one place. Tried a few options. \-- LiteLLM: Python, works fine at low volume. At \~300 req/min the latency overhead was adding up. About 8ms per request. \--Custom nginx+lua: Got basic routing working but the failover and budget logic was becoming its own project. Bifrost (OSS - [https://git.new/bifrost](https://git.new/bifrost) ): What I ended up with. Go binary, Docker image, web UI for config. 11-15 µs overhead per request only. Single endpoint, all providers behind it. The semantic caching is what actually saves money. Uses Weaviate for vector similarity. If two users ask roughly the same thing, the second one gets a cached response. Direct hits cost zero tokens. Runs on a single $10/mo VPS alongside our other stuff. Hasn't been a resource hog. Config is a JSON file, no weird DSLs or YAML hell. Honestly the main thing I'd want improved is better docs around the Weaviate setup. Took some trial and error.