Desktop/Browser tasks automater

A local browser and desktop operator for internships, scholarships, Docs, Drive, Slides, D2L, and daily desktop actions.

1. Browser sessions are now reused through a stable session_id, so repeated commands stay in the same automation browser.
2. The default planner profile is now qwen2_5_7b for stronger text first planning.
3. The browser worker now supports direct text clicks, CSS clicks, typing, key presses, waiting for text, tab listing, tab switching, tab closing, and basic challenge detection.
4. The desktop worker now supports PowerShell commands, file search, file open, file move, clipboard copy, paste, and Notepad launch.
5. The engine now routes mixed tasks instead of only internships and scholarships.
6. The voice frontend now has a real engine bridge that can transcribe a wav file, submit the task, and speak a summary response.

1. Create a virtual environment.
2. Install requirements.
3. Run playwright install chrome.
4. Start the model with scripts/start_model.ps1 -Profile qwen2_5_7b.
5. Start the browser service.
6. Start the desktop service.
7. Start the engine service.
8. Optionally start the Telegram bridge.

1. The browser stays visible on purpose.
2. The same session_id now maps to the same persistent Playwright profile directory.
3. Google Docs and Drive are currently browser first in this codebase. API first integration can be layered on later.
4. Challenge detection is included, but challenge completion is still a human handoff step.