[Feature Request] Persistent System Instructions

### **Is your feature request related to a problem?**
Yes. Currently, users with specific requirements (e.g., specific coding styles, formatting rules like LaTeX, or professional personas) must manually re-input these instructions at the start of every new chat. This creates "prompt fatigue" and lead to inconsistent model behavior across different threads.

### **Describe the solution you'd like**
I would like to see a feature similar to "Instructions for Gemini" or ChatGPT’s "Custom Instructions" implemented in the DeepSeek Web App.

**Key Features:**
* **Persistent Global Field:** A dedicated section in the Web App settings where users can define "How DeepSeek should respond" and "What DeepSeek should know about me."
* **System Prompt Injection:** These instructions should be automatically included as a `system` message in every new chat session.
* **Toggle Capability:** An option to enable/disable these instructions for specific chats if a "neutral" response is needed.

### **Describe alternatives you've considered**
* **Manual Copy-Paste:** This is inefficient and consumes significant user time.
* **Browser Extensions:** These are unreliable and often break when the Web App UI is updated.
* **API Usage:** While this is possible via the DeepSeek API, it is not accessible to the mainstream Web App audience, limiting the UX for power users.

### **Additional Context**
Providing this feature would significantly improve the "stickiness" of the DeepSeek Web App for professionals, researchers, and developers who have non-negotiable requirements for output structure and tone.