feat: add MiniMax provider preset with MiniMax-M3 and MiniMax-M2.7 models

[octo-patch]

Reason: Add a first-class MiniMax provider preset on the existing OpenAI-compatible transport.

Adds the MiniMax provider to the provider registry so it is selectable in the provider manager and model picker, alongside its two current text models.

Changes:

• Register the minimax provider id, display name, and default base URL in extension/src/api/providers/constants.ts (global endpoint https://api.minimax.io/v1).
• Add extension/src/api/providers/config/minimax.ts with model metadata for MiniMax-M3 (1,000,000 context window, text/image/video input, adaptive thinking, prompt cache) and MiniMax-M2.7 (204,800 context window, text input, always-on thinking, prompt cache), including per-million-token pricing and cache read/write prices.
• Wire the provider into providerConfigs in extension/src/api/providers/config/index.ts so its models appear in listModels/currentModelInfo.
• Route the minimax provider through the OpenAI-compatible SDK transport in extension/src/api/providers/custom-provider.ts, using the global base URL by default and allowing an optional CN base URL override.
• Add a MiniMaxSettings type (providerId, apiKey, optional baseUrl) to the ProviderSettings union in extension/src/api/providers/types.ts.
• Add a settings form for the MiniMax provider (API key + optional base URL, with the global and CN endpoint hints) in extension/webview-ui-vite/src/components/settings-view/preferences/provider-manager.tsx.

Checks:

• tsc --noEmit (extension) — passes.
• eslint src (extension) — passes.
• tsc -b --force (webview-ui-vite) — passes.

[octo-patch]

Updated the MiniMax provider to expose the global and China OpenAI-compatible and Anthropic-compatible endpoints, route each format through the matching transport, and record the current input and thinking metadata. I ran the extension typecheck and lint, the webview TypeScript build, and targeted provider assertions.