feat: add MiniMax as LLM provider (llm-chain-minimax crate)

[octo-patch]

Summary

Adds a new llm-chain-minimax crate that integrates MiniMax LLM models via their OpenAI-compatible API, following the same architecture pattern as llm-chain-openai.

What's included

• New crate llm-chain-minimax — Full Executor trait implementation using async-openai with MiniMax's API endpoint (https://api.minimax.io/v1)
• Model support — MiniMax-M2.7, M2.7-highspeed, M2.5, M2.5-highspeed with context window metadata (up to 1M tokens)
• Think-tag stripping — Automatically strips <think>...</think> reasoning tags from M2.5+ model responses
• executor!(minimax) macro — Integrated into the existing executor macro system
• Environment-based config — MINIMAX_API_KEY and optional MINIMAX_API_BASE_URL
• 16 unit tests + 3 integration tests — Covering model parsing, tokenization, context sizes, and live API calls
• README updates — Added MiniMax to the features list and getting started guide

Supported models

Model	Context Window
MiniMax-M2.7	1M tokens
MiniMax-M2.7-highspeed	1M tokens
MiniMax-M2.5	1M tokens
MiniMax-M2.5-highspeed	204K tokens

Files changed (12 files, 934 additions)

• crates/llm-chain-minimax/ — New crate (Cargo.toml, README, src/, tests/, examples/)
• crates/llm-chain/src/executor.rs — Added minimax variant to executor! macro
• docs/README.md — Added MiniMax to features and getting started sections

Test plan

• All 16 unit tests pass (cargo test -p llm-chain-minimax)
• All 3 integration tests pass with live MiniMax API (--ignored flag)
• Doc-tests pass
• Crate builds without errors
• Verify existing llm-chain-openai tests still pass (no modifications to existing crate logic)