Add DeepSeek to SUPPORTED_PROVIDERS

[winsonluk]

This change adds deepseek:deepseek-chat, which is the latest DeepSeek-V3 model (released December 26, 2024).

DeepSeek doesn't have its own LangChain, so official DeepSeek documentation suggests using OpenAI's LangChain (https://api-docs.deepseek.com/faq), which this change implements.

Confirmed working with FAST_LLM="deepseek:deepseek-chat", SMART_LLM="deepseek:deepseek-chat", and STRATEGIC_LLM="deepseek:deepseek-chat" when DEEPSEEK_API_KEY system variable is set.

[winsonluk]

@assafelovic this adds DeepSeek as an additional provider since their DeepSeek-V3 benchmarks are pretty impressive

[assafelovic]

Love this! How's the model performance?

[winsonluk]

Love this! How's the model performance?

@assafelovic so far so good! I'm integrating it into https://ithy.com

It's a perfect candidate for FAST_LLM (faster and cheaper than 4o-mini / 3.5 Haiku / 1.5 Flash). When used as SMART_LLM or STRATEGIC_LLM, performance even exceeds most frontier models (though it starts to underperform when context tokens get very large).

My only concern is that they don't have published rate limits, but I haven't gotten any 429s so far 🤞