feat: send video as multiple frames (1 FPS, max 64) to llama-server

fix: extract video frame as JPEG before sending to llama-server

fix: pass --mmproj to llama-server for vision/video support

fix: make uploaded videos playable in chat bubbles

fix: add video upload support and file loading feedback to simple UI

feat: add simple chat mode (llama-server only, no Docker/backend/frontend)

fix: use taskkill for process management on Windows

fix: use rglob to recursively check for files in support model dirs

fix: check support model dirs are non-empty before skipping download

fix: add direct llama-server path to binary search candidates

fix: set PYTHONIOENCODING=utf-8 for service subprocesses on Windows

fix: handle WinError 87 on print(flush=True) in Windows console

fix: generate SSL cert without openssl on Windows

fix: check for support model dirs before skipping model download

fix: UTF-8 encoding for pyproject.toml, robust pnpm resolution