Chat messages intermittently trigger ComfyUI image generation instead of LLM responses (Advanced Chat, GGUF models)

## Describe the bug

During normal text‑only chat in Advanced Chat mode, the Playground intermittently routes user messages into the ComfyUI image generation backend instead of the LLM backend. This causes an image to be generated (using the last active image preset) instead of producing a text response.

The issue occurs even when:

- No image generation was requested
- The user is in the Chat tab
- Tools are disabled
- The active model is a standard GGUF LLM (e.g., Llama 3.1, Mistral 7B, etc.)

Logs show the Playground incorrectly calling:
comfyUI.execute {workflow: 'Draft Image', prompt: 'Hello, I'}

This happens after switching between GGUF models or switching between Chat/Image Gen modes, suggesting a UI state‑corruption issue.

## To Reproduce

The issue is intermittent, but the following sequence reliably triggered it:

1. Open Advanced Chat mode.
2. Load a GGUF model (e.g., Llama‑3.1‑8B‑Instruct or Mistral‑7B‑Instruct‑v0.3).
3. Switch to another GGUF model without clearing the chat session.
4. Switch between Chat ↔ Image Gen tabs at least once.
5. Type a normal chat message such as “hi” or “hello”.

**Actual behavior:**
Instead of sending the message to the LLM, the Playground triggers ComfyUI:
### comfyUI.execute {workflow: 'Draft Image', prompt: 'Hello, I'}
### presets store activePresetWithVariant Pro Image
### computing backend comfyui

An image is generated using the last active image preset.

## Expected behavior

Chat messages in Chat mode should always be routed to the selected LLM backend (llama.cpp, OpenVINO, etc.), regardless of:

- model switching
- preset switching
- previous use of Image Gen
- tool‑calling capabilities

Image generation should only occur when explicitly requested.

## Screenshots

## Environment (please complete the following information):

OS: Windows 11
GPU: Intel Arc A750
CPU: Intel i3‑10100
RAM: 20 GB
AI Playground Version: 3.0.2b

## Additional context

- Issue disappears temporarily after clearing chat logs and restarting the Playground.
- Not tied to any specific GGUF model.
- Appears only after switching models or switching between Chat/Image Gen tabs.
- This is separate from the conversation‑roles template error (issue #420).
- Logs strongly suggest the Chat UI state becomes desynchronized, causing the backend router to select ComfyUI instead of the LLM.