add initial support for Fireworks provider

[ofou]

Adds support for the Fireworks AI platform as a new provider option, specifically integrating the deepseek-r1 model via the Fireworks. This addition expands our available providers options, giving users access to more powerful reasoning capabilities, particularly with DeepSeek's large context window of 160k tokens, but faster with Fireworks.

The implementation follows existing provider architecture pattern while adding Fireworks-specific adaptations:

• Added new FireworksHandler class that implements the ApiHandler interface
• Created a custom stream handling mechanism for Fireworks SSE responses with proper error handling
• Added special handling for DeepSeek models which may require different message formatting and parameters for temperature (recommended)
• Defined Fireworks model configuration in shared API types
• Added retry logic to handle potential API failures
• Integrated the handler into our existing provider selection system
• Added appropriate configuration options including the Fireworks API key

The implementation uses the OpenAI compatible library with Fireworks.

[ofou]

It's only supporting deepseek-v3 and deepseek-r1. I tried to follow former Provider implementations

If you need a Fireworks key to test this, ping me on slack :)

[janpaul123]

I haven't looked too deeply yet, but why is this file so much longer than most of the other tests in /src/api/providers/__tests__? Just wondering if there are some unnecessary tests.

[ofou]

No one has ever claimed to have enough tests... but Kilocode helped me with this 😉. I also borrowed some inspiration from other implementations, but there are a few special cases, such as the </thinking> tokens from DeepSeek's models in the response

[janpaul123]

Overall looks good to me know, aside from some git hygiene. I'll let @kevinvandijk also have a look, and maybe he can show you how to do better git hygiene, because he's good at that. Overall, nice, let's ship this soon!