feat: add MiniMax as first-class LLM provider (M3 default, chat + embedding)

[octo-patch]

Summary

Add MiniMax as a fully supported LLM provider alongside OpenAI and Doubao, covering both chat completion (via OpenAI-compatible API) and embedding generation (via MiniMax native API). The default chat model is MiniMax-M3 (512K context window, 128K max output, image input).

Changes

Backend (opencontext/llm/llm_client.py):

• Add MINIMAX enum to LLMProvider
• Implement _minimax_embedding() / _minimax_embedding_async() for MiniMax native embedding API (embo-01 model, non-OpenAI-compatible format using texts/type/vectors)
• Fix _request_embedding() to avoid NameError on undefined response variable in the MiniMax code path
• Fix _request_embedding_async() to route MiniMax to native API instead of falling through to incompatible OpenAI SDK
• Add MiniMax embedding validation in validate()
• Add MiniMax error codes (invalid_api_key, insufficient_balance) to _extract_error_summary()

Frontend (settings/constants.tsx, settings.tsx):

• Add MiniMax to ModelTypeList with M3 (default), M2.7 and M2.7-highspeed chat models
• Add embo-01 embedding model and api.minimax.io base URL presets
• Add MiniMax provider icon (minimax.svg)
• Wire up API key link, base URL routing, and form rendering

Documentation (README.md, README_zh.md):

• List MiniMax as supported provider in Quick Start and Backend Architecture sections
• Add minimax option in config.yaml examples

Tests (22 unit + 3 integration):

• Provider enum, client init, chat completion, streaming, thinking mode
• Default model is now MiniMax-M3; M2.7-highspeed continues to be covered
• Embedding: native API format, auth headers, error handling, dimension truncation
• Validation: chat and embedding success/failure paths
• Error handling: MiniMax-specific error code extraction
• Integration tests with real MiniMax API (requires MINIMAX_API_KEY)

Test Plan

• All 22 unit tests pass with M3 as default
• 2/3 integration tests pass (embedding test may hit MiniMax rate limits)
• Frontend builds successfully with new provider
• Manual verification with MiniMax API key in the Settings UI

MiniMax Models

Model	Context	Max Output	Use Case
MiniMax-M3 (default)	512K tokens	128K tokens	Chat / VLM (image input)
MiniMax-M2.7	192K tokens	—	Previous-gen chat
MiniMax-M2.7-highspeed	192K tokens	—	Previous-gen low-latency chat
embo-01	—	—	Embedding (1536 dims)