feat: add MiniMax as first-class LLM provider (M3 default)

[octo-patch]

Summary

Add MiniMax AI as a first-class LLM provider with MiniMax-M3 as the default model (alongside M2.7 series).

• MiniMax offers an OpenAI-compatible API
• MiniMax-M3 (default) — 512K context, up to 128K output, image input support
• MiniMax-M2.7 — 192K context (previous generation)
• MiniMax-M2.7-highspeed — 192K context, lower latency variant
• New MiniMaxChat class extending BaseChatModel with temperature clamping, think-tag stripping, and cost tracking
• Registered minimax, MiniMax-M3, MiniMax-M2.7, MiniMax-M2.7-highspeed in llm_registry

Changes

File	Description
agentverse/llms/minimax.py	New MiniMax LLM provider module (M3 default)
agentverse/llms/__init__.py	Import MiniMaxChat
README.md	MiniMax environment variables and config docs
tests/test_minimax_unit.py	43 unit tests
tests/test_minimax_integration.py	3 integration tests

Pricing (per 1M tokens)

Model	Input	Output
MiniMax-M3 (default)	$0.60	$2.40
MiniMax-M2.7	$0.80	$3.20
MiniMax-M2.7-highspeed	$0.80	$3.20

Usage

export MINIMAX_API_KEY="your_key"

# In task config YAML
llm_type: minimax
model: MiniMax-M3

Test plan

• 43 unit tests passing (mocked API, registry, cost tracking, think-tag stripping, temperature clamping)
• 3 integration tests passing (basic generation, history, async) using MiniMax-M3
• Verify existing OpenAI tests still pass