wip <> WASM-tested baseline models yesterday / doing webgpu binds

required pinned ggml patch & OMP,SIMD,etc. tweaks to make all models efficient, this work meshes well with other builds been doing with all-way sherpa onnx etc.