feat: add Kimi K2 thinking model to Fireworks AI provider

[roomote]

This PR attempts to address Issue #9201 by adding support for the Kimi K2 thinking model to the Fireworks AI provider.

Changes

• Added accounts/fireworks/models/kimi-k2-thinking to the FireworksModelId type
• Added model configuration for kimi-k2-thinking with appropriate parameters matching the Moonshot provider implementation
• Added comprehensive test coverage for the new model

Testing

• All existing tests pass
• Added new test case specifically for the Kimi K2 thinking model
• Verified model configuration matches expected values

Fixes #9201

Feedback and guidance are welcome!

---

Important

Add kimi-k2-thinking model to Fireworks AI provider with configuration and tests.

• Behavior:
  • Add accounts/fireworks/models/kimi-k2-thinking to FireworksModelId in fireworks.ts.
  • Configure kimi-k2-thinking model in fireworksModels with parameters like maxTokens, contextWindow, supportsPromptCache, etc.
• Testing:
  • Add test case in fireworks.spec.ts to verify kimi-k2-thinking model configuration.
  • Ensure all existing tests pass and new tests cover the new model's functionality.

^{This description was created by for 811c732. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Rooviewer   See task on Roo Cloud

Review complete. Found 1 issue that needs attention.

• Temperature default conflict: kimi-k2-thinking model specifies defaultTemperature 1.0 but FireworksHandler overrides with 0.5

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[roomote]

The kimi-k2-thinking model specifies defaultTemperature: 1.0 and supportsTemperature: true, but the FireworksHandler constructor sets a blanket defaultTemperature: 0.5 for all models. This means the model-specific default temperature of 1.0 will be overridden by the handler's 0.5 default, potentially affecting the model's reasoning behavior since temperature is a critical parameter for thinking models.

_{Fix it with Roo Code or mention @roomote and request a fix.}

[mmealman]

I tried working with Kimi K2 Thinking a few days ago with Fireworks and tool calls were failing. However in testing this PR with Fireworks and Kimi K2 Thinking everything is working fine now. So either the interleave thinking issue at Fireworks was fixed or some other change in Roo also fixed the interleave tool call problems.

Looking forward to this being merged into main so the rest of my team can access it.