[FEATURE]:Native provider support and specialized system prompts for domestic LLMs (DeepSeek, Qwen, GLM)

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

🎯 Problem Statement

Currently, OpenCode provides excellent built-in, per-model optimized system prompts for Western models like Anthropic's Claude, OpenAI's GPT, and Google's Gemini. However, to use top-tier domestic models such as DeepSeek-R1/V4, Qwen-2.5-Coder, or GLM-4, users have to rely on the generic openai-compatible adapter.

While these models are highly capable, a one-size-fits-all generic prompt often causes suboptimal performance. For instance:

• DeepSeek-R1 cannot fully leverage its <think> tag workflows inside OpenCode's agent tool loop.
• Qwen-Coder occasionally generates slight format discrepancies when producing structural JSON/XML Diff patches under general prompts, leading to tool-calling failures.

✨ Proposed Solution

I propose introducing native provider drivers and specialized system prompts (.txt templates) tailored for these leading domestic LLMs:

1. Dedicated System Prompts: Add specific routing for deepseek.txt and qwen.txt inside the core prompt manager, optimizing tool-calling constraints and Diff patch consistency for them.
2. DeepSeek Reasoner Adaptation: Optimize the prompt to explicitly instruct DeepSeek models to cleanly handle thinking blocks without breaking the terminal streaming or UI parsing.
3. First-class Config Support: Allow users to specify "provider": "deepseek" or "provider": "dashscope" (Qwen) natively in opencode.jsonc, rather than nesting them under openai-compatible.