Add MiniMax as LLM provider backend

[octo-patch]

Summary

Adds MiniMax as a first-class REST LLM provider backend for spacy-llm, alongside OpenAI, Anthropic, Cohere, and PaLM.

MiniMax offers an OpenAI-compatible chat completions API with models like MiniMax-M2.5 (204K context) and MiniMax-M2.7 (1M context), making it a compelling alternative for structured NLP pipelines.

Changes

• New provider module spacy_llm/models/rest/minimax/ with model, registry, and init files
• Registry entry spacy.MiniMax.v1 for seamless spaCy config integration
• Think-tag stripping for reasoning model output
• Context length definitions for M2.5, M2.5-highspeed, M2.7, M2.7-highspeed
• Test suite with unit tests (context lengths, endpoints) and integration tests (API response validation, error handling)
• Usage example config for text classification with MiniMax
• README updated to list MiniMax as a supported provider

Usage

Python

import spacy

nlp = spacy.blank("en")
nlp.add_pipe(
    "llm",
    config={
        "model": {
            "@llm_models": "spacy.MiniMax.v1",
            "name": "MiniMax-M2.5",
            "config": {"temperature": 0.0},
        },
        "task": {
            "@llm_tasks": "spacy.TextCat.v2",
            "labels": ["POSITIVE", "NEGATIVE"],
        },
    },
)
doc = nlp("This is a great product!")
print(doc.cats)

Config file

[components.llm.model]
@llm_models = "spacy.MiniMax.v1"
name = "MiniMax-M2.5"
config = {"temperature": 0.0}

Supported Models

Model	Context Length
MiniMax-M2.7	1,048,576
MiniMax-M2.7-highspeed	1,048,576
MiniMax-M2.5	204,800
MiniMax-M2.5-highspeed	204,800

Test Plan

• Unit tests pass: context length definitions, endpoint constants
• Integration tests pass: API response validation with M2.5 and M2.5-highspeed
• Error handling tests pass: unsupported model, invalid config
• Registry integration verified: spacy.MiniMax.v1 resolves correctly
• Existing tests unaffected