Companion to the LLM_JSON_MODE feature: documents how to point MiroFish
at LM Studio, Ollama, and llama.cpp server, including the
LLM_JSON_MODE=none requirement, recommended context window for
ontology generation, memory/throughput expectations, and Apple Silicon
caveats (the macOS Tahoe + Mx Metal shader bug in Ollama 0.21).
Adds a one-line callout in README.md pointing to the new doc, so the
main quickstart stays focused on the cloud-LLM happy path.
d42fce93c6