fix(ui): remove experimental webgpu tab and switch semantic cache back to stable numpy arrays

feat: precompute semantic embeddings to drastically reduce startup time and prevent Gradio timeouts

feat: complete local embedding search with i-dot-ai HF model