Desktop App: Add setting to disable auto-scroll during streaming generation (scroll locking)

Feature request: Add a setting to disable automatic scrolling during token streaming/generation in the Desktop App Console.

Problem: When streaming generation is enabled, the Desktop App automatically scrolls to follow new tokens in real time. There is no toggle to lock the viewport so the user can read/scroll at their own pace while the model is still generating.

Current state: The only workaround is to wait for completion (then manually scroll back) or type to interrupt generation. No config option (config.json or UI Settings) exists for this behavior.

Requested behavior: Add either:

A UI toggle: "Lock scroll during generation" (default: off / follow)
A config option: auto_scroll_on_update: true|false (applied per channel or globally)
Environment:

QwenPaw version: 2.1.0 (Desktop App, Tauri)
OS: macOS (Apple Silicon M5 Pro)
oMLX backend, local models
Which op

Start by locating the Desktop App Console's streaming-generation scroll behavior and the UI Settings or config.json paths mentioned in the issue. Determine how a per-channel or global setting should be represented and persisted. Done means a default-off lock-scroll option prevents automatic scrolling during generation while preserving the existing follow behavior when disabled.

Vom Indexierungsmodell aus dem Issue-Text verfasst.