Complete MiniMax provider transport and model metadata

[octo-patch]

Reason: Complete MiniMax provider support with current regional Messages transport and provider-level model metadata.

Changes

• Add regional Chat Completions and Messages transport presets and request handling.
• Record context windows, token pricing, input modalities, and thinking modes for MiniMax-M3 and MiniMax-M2.7.
• Add focused coverage for provider resolution, transport requests, and the complete model metadata registry.
• Document the provider metadata exposed by PROVIDER_MODELS.

Verification

• python3 -m py_compile molecode/llm.py tests/test_llm_provider_presets.py
• python3 - with offline metadata, endpoint, request, and response assertions
• git diff --check

[octo-patch]

I pushed a follow-up commit that updates the MiniMax Anthropic Messages request to use the documented API-key header and avoids sending the incompatible zero default temperature. I ran Python bytecode compilation, git diff --check, and a focused mocked transport check.