Refresh MiniMax presets for M3

[octo-patch]

Reason: llmprovider/presets.yml is a go:embed static provider preset directory, with existing MiniMax presets minimax_intl and minimax_cn but models only up to M2.7.

Summary

• Add MiniMax-M3 and MiniMax-M3-thinking to both international and CN MiniMax presets.
• Refresh MiniMax OpenAI-compatible base URLs to the current global and CN endpoints.
• Add international and CN Anthropic-compatible presets with MiniMax-M3 and MiniMax-M2.7.
• Align MiniMax-M3 and MiniMax-M2.7 context and output limits with the current API specifications.

Checks

• git diff --check
• go test llmprovider/presets.go llmprovider/types.go
• Focused Go assertions for all four MiniMax preset variants
• Request-path capture for /v1/chat/completions and /anthropic/v1/messages
• go test ./llmprovider (fails before package compilation because the repository-local ../gou, ../kun, ../xun, and ../v8go replacements are absent from this checkout)

[octo-patch]

Updated both MiniMax presets with the current context limits, multimodal capability, and thinking request shape. I ran git diff --check, structured YAML assertions, and go test llmprovider/presets.go llmprovider/types.go.

[octo-patch]

Updated the MiniMax presets to add international and CN Anthropic-compatible entries and align the MiniMax-M3 and MiniMax-M2.7 token limits with the current API specifications. I ran git diff --check, the focused Go preset loader check, structured four-preset assertions, and request-path capture for both compatible protocols.