Add DeepSeek provider (OpenAI-compatible API)

[djthorpe]

Summary

Implement a DeepSeek provider (pkg/provider/deepseek). DeepSeek's API is OpenAI-compatible, making this a straightforward implementation.

Requirements

• Implement llm.Client interface for the DeepSeek API (https://api.deepseek.com)
• API key via DEEPSEEK_API_KEY environment variable
• Support streaming and non-streaming chat completions
• Support tool/function calling
• Support thinking/reasoning (DeepSeek-R1 uses <think> blocks)
• Register provider in the server command alongside existing providers

Models

• DeepSeek-V3 (chat)
• DeepSeek-R1 (reasoning)
• DeepSeek-Coder (code-focused)

Notes

• The API follows OpenAI's chat completions format, so much of the implementation can mirror or share code with a future OpenAI provider
• There is existing code in etc/_old/deepseek/ that may be useful as reference
• DeepSeek-R1 reasoning output appears in the reasoning_content field of the response, which needs to be mapped to the thinking content block model

Motivation

Quick win — OpenAI-compatible API means minimal new code. DeepSeek-R1 is a popular reasoning model and adds another thinking-capable provider alongside Gemini and Anthropic

• Embedding models (text-embedding-3-small, text-embedding-3-large)

Motivation

OpenAI is a major LLM provider and its multi-modal output capabilities (images, audio) will drive the content model to support rich responses across all providers, improving the overall architecture.