Refresh MiniMax defaults and endpoints

[octo-patch]

Summary

• Default MiniMax provider selections to MiniMax-M3 in the model wizard.
• Expose the current global and domestic OpenAI-compatible and Anthropic-compatible endpoints.
• Use SDK-compatible /anthropic base URLs so the client constructs exactly one /v1/messages path.
• Use protocol-compatible thinking defaults: adaptive for OpenAI-compatible MiniMax-M3, the disabled default for Anthropic-compatible MiniMax-M3, and the API's built-in always-on behavior for MiniMax-M2.x.
• Add request-path coverage for both regions, both protocols, and the MiniMax-M3 and MiniMax-M2.7 model IDs.

Testing

• bun run check (233 tests)
• bun test src/cli/__tests__/model-providers.test.ts (10 tests)
• git diff --check

[octo-patch]

Updated the PR to use API-compatible MiniMax thinking defaults: MiniMax-M3 now sends adaptive, while MiniMax-M2.x relies on its built-in always-on behavior. I also added request-path coverage for both regions and protocols. bun run check passes with 233 tests, and the focused provider suite passes with 10 tests.

[octo-patch]

Updated the runtime defaults so MiniMax-M3 now follows the configured protocol: adaptive for OpenAI-compatible entries and disabled for Anthropic-compatible entries. bun run check (233 tests), the focused provider suite (10 tests), and git diff --check pass.