Add MiniMax as LLM provider

[octo-patch]

Summary

Adds MiniMax AI as a new LLM provider for autolabel, enabling users to label datasets using MiniMax's cloud models via the OpenAI-compatible API.

Supported Models

• MiniMax-M2.7 (latest, 1M context window)
• MiniMax-M2.5 (204K context)
• MiniMax-M2.5-highspeed (204K context, lower latency)

Changes

• New file: src/autolabel/models/minimax.py — Full provider with sync/async labeling, retry logic, think-tag stripping, temperature clamping [0,1], and per-model cost tracking
• Modified: src/autolabel/schema.py — Added MINIMAX = "minimax" to ModelProvider enum
• Modified: src/autolabel/models/__init__.py — Registered MiniMaxLLM in MODEL_REGISTRY
• Modified: README.md — Added MiniMax to the list of supported LLM providers
• New: 25 unit tests + 3 integration tests with 8 config files

Usage

{
  "model": {
    "provider": "minimax",
    "name": "MiniMax-M2.7",
    "params": {
      "temperature": 0.1,
      "max_tokens": 1000
    }
  }
}

Requires MINIMAX_API_KEY environment variable. Get an API key at https://platform.minimaxi.com.

Test Plan

• 25 unit tests passing (initialization, labeling, cost calculation, think-tag stripping, error handling, async, registry)
• 3 integration tests passing with real MiniMax API (sync label, M2.5-highspeed label, async label)
• Existing tests unaffected

14 files changed, 845 additions, 1 deletion