Wondering what the possibilities are for 32bit memory allocation. It appears this requires, 64bit memory pipeline to load models. Thoughts? Point is our team would like to target "wasm32" as a build target.