Add MiniMax as first-class LLM provider

[octo-patch]

Summary

Add MiniMax as a first-class LLM provider for document extraction, supporting MiniMax-M2.7 and MiniMax-M2.7-highspeed models (up to 1M context window).

Users can simply use the minimax/ model prefix. The integration auto-detects MiniMax models, rewrites the model string for litellm OpenAI-compatible provider, and configures the API base URL and key automatically from environment variables.

Usage

import os
from extract_thinker import Extractor, DocumentLoaderPyPdf, Contract

os.environ["MINIMAX_API_KEY"] = "your-key"

class InvoiceContract(Contract):
    invoice_number: str
    invoice_date: str

extractor = Extractor()
extractor.load_document_loader(DocumentLoaderPyPdf())
extractor.load_llm("minimax/MiniMax-M2.7")

result = extractor.extract("invoice.pdf", InvoiceContract)

Changes

• extract_thinker/llm.py - Add api_base/api_key constructor params, MiniMax auto-detection via minimax/ prefix, temperature clamping to [0, 1], extra params passed through all request paths
• extract_thinker/utils.py - Add is_minimax_model(), resolve_minimax_model() helpers and MINIMAX_API_BASE/MINIMAX_MODELS constants
• extract_thinker/global_models.py - Add get_minimax_model() and get_minimax_highspeed_model() convenience functions
• README.md - Add MiniMax to supported providers list with usage example
• tests/test_minimax.py - 29 unit tests + 3 integration tests covering model detection, temperature clamping, parameter wiring, and real API calls

Test plan

• 29 unit tests pass (model detection, resolve, temperature clamping, extra params, global models)
• 3 integration tests pass with real MiniMax API (raw completion, structured extraction via Extractor, highspeed variant)
• Existing test_llm_backends.py tests unaffected (pre-existing failures not related to this change)
• CI pipeline passes