Add MiniMax as alternative LLM provider (default model: M3)

[octo-patch]

Summary

• Add multi-provider LLM factory (utils/llm_provider.py) supporting OpenAI (default) and MiniMax via LLM_PROVIDER env var
• Default MiniMax chat model is MiniMax-M3 (512K context, 128K max output, image input)
• Keep MiniMax-M2.7 and MiniMax-M2.7-highspeed available as overrides
• Add MiniMaxEmbeddings class for MiniMax embo-01 (1536 dims) native API integration
• Add bonus notebook [6]_rag_with_minimax.ipynb demonstrating full RAG pipeline with MiniMax M3 + embo-01
• Update .env.example with MiniMax configuration options
• Update README.md with provider comparison table and quick-start guide

Key Features

• Zero code changes: Switch providers by setting LLM_PROVIDER=minimax in .env
• Latest model default: New requests default to MiniMax-M3 — the most recent MiniMax release
• Temperature clamping: Automatically clamps temperature to MiniMax-accepted [0, 1] range
• Native embedding support: MiniMaxEmbeddings wraps the proprietary embo-01 API (uses texts/type fields, not OpenAI-compatible)
• Drop-in replacement: get_chat_model() returns a standard LangChain ChatOpenAI instance that works with all existing chains

Models

Model	Notes
MiniMax-M3 (default)	512K context window, 128K max output, image input supported
MiniMax-M2.7	Previous-generation model
MiniMax-M2.7-highspeed	Previous-generation low-latency variant

Test Plan

• 44 unit tests covering provider resolution, temperature clamping, factory methods, embedding API calls
• 6 integration tests (3 chat + 3 embedding) with real MiniMax API
• All unit tests pass locally
• Chat integration tests pass with live API
• Embedding integration tests pass (subject to rate limits)