Friendly error message when ZeroGPU daily quota is exhausted

Increase Qwen GPU duration budget: first call moves 40GB to VRAM

Fix ZeroGPU: load Qwen at module level in main thread, not background thread

Add Qwen-Edit-2509 Upscale LoRA as a second tab (background-loaded)

Suppress noisy ZeroGPU event-loop tracebacks; use HF_TOKEN for weight downloads

Add multi-model upscaler on ZeroGPU

initial commit