Deploy qmd-web

feat: manual generation for no-KV-cache model, fix blend scoring, add coffee doc

fix: add timeout for expansion model loading, rebuild

fix: use self-contained ONNX models, sequential loading, Vite config

fix: model loading for browser compatibility

fix: use AutoModelForCausalLM for reranker (yes/no token scoring)

add model loading layer for Transformers.js WebGPU pipelines