docs: document LLM model configuration for MCP tools

[gaodan-fang]

Context

PR #238 surfaced that store_user_facts fails when the environment's default fact-extraction model is unavailable or misconfigured. The failure occurs at the litellm.completion() call, not in the persistence path, making it hard to diagnose without knowing the configuration surface.

Problem

The three LLM-backed features — guidelines generation, conflict resolution, and fact extraction — each rely on a model setting in altk_evolve/config/llm.py (LLMSettings), but the required environment variables and their fallback chain are not documented:

Setting	Env var	Fallback
guidelines_model	EVOLVE_GUIDELINES_MODEL	EVOLVE_TIPS_MODEL → EVOLVE_MODEL_NAME → gpt-4o
conflict_resolution_model	EVOLVE_CONFLICT_RESOLUTION_MODEL	EVOLVE_MODEL_NAME → gpt-4o
fact_extraction_model	EVOLVE_FACT_EXTRACTION_MODEL	EVOLVE_MODEL_NAME → gpt-4o
custom_llm_provider	EVOLVE_CUSTOM_LLM_PROVIDER	auto-detects openai if OPENAI_BASE_URL or OPENAI_API_KEY is set

Suggested scope

• Add a configuration reference (README section or dedicated doc) covering these env vars.
• Consider surfacing a clearer error when a model is not reachable, rather than the generic Failed to parse extracted facts response after 3 retries.

Ref

• feat(mcp): user facts tools, SSE transport hardening, and warmup #238 (comment)

[gaodan-fang]

Closing — this is already documented in docs/guides/configuration.md, which covers all LLM model env vars, fallback precedence, and the deprecated EVOLVE_TIPS_MODEL alias.