Faster WebGPU sampling: fp16 compute, GPU-resident steps, non-blocking thinker preview

Preview 002 and 003 (512 px), version selector, Show thinker, AI-output marking

Count every model load, like from_pretrained

Read the model config once per browser for the download counter

Count first-time model loads on the Logolabs/agate-preview-001 download counter

Agate WebGPU: static in-browser demo with ONNX models