feat: Add Kimi K2 0905 model to Groq, Moonshot, and Fireworks providers

[daniel-lxs]

Summary

Adds the new Kimi K2 0905 model version to multiple providers with extended context window support.

Changes

• Added Kimi K2 0905 model to Groq, Moonshot, and Fireworks providers
• Updated default models to use the new Kimi K2 0905 version
• Extended context window from 128K to 256K tokens
• Added prompt caching support where applicable

Improvements

• Better agentic coding with improved accuracy and generalization
• Enhanced frontend coding capabilities
• Extended context length for better long-horizon support

All tests updated and passing.

---

Important

Adds Kimi K2 0905 model to Groq, Moonshot, and Fireworks providers with extended context window and prompt caching support.

• Behavior:
  • Adds Kimi K2 0905 model to fireworks.ts, groq.ts, and moonshot.ts with extended context window to 256K tokens.
  • Updates default model to Kimi K2 0905 in fireworks.ts, groq.ts, and moonshot.ts.
  • Adds prompt caching support to Kimi K2 0905 model.
• Tests:
  • Updates moonshot.spec.ts to reflect new model's maxTokens and contextWindow values.
  • Ensures supportsPromptCache is true for Kimi K2 0905 model in tests.

^{This description was created by for b7f77c7. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Thank you for your contribution! I've reviewed the changes and found several issues that need attention before this can be merged.

[roomote]

Is this intentional? The PR description mentions adding Kimi K2 0905 model to existing providers, but this introduces an entirely new 'anthropic-compatible' provider. This is a significant architectural change that should be mentioned in the PR title and description. Could we consider splitting this into a separate PR to keep changes focused?

[roomote]

This class appears to be nearly identical to AnthropicHandler. Could we reduce code duplication by extending AnthropicHandler instead?

[roomote]

The getModel() method doesn't handle the case when a user provides a custom model ID that isn't in the anthropicModels list. Should we add validation or fallback logic here to handle custom models properly with anthropicCustomModelInfo?

[roomote]

The model dropdown is hidden for anthropic-compatible provider, but there's no alternative way to select from the predefined anthropicModels list. Users can only manually type model IDs. Is this the intended UX? Perhaps we should show the dropdown but also allow custom input?

[roomote]

These strings should use the translation system for consistency:

[roomote]

This label should also use the translation system:

[roomote]

This label should use the translation system: