I think this is definitely an improvement over the wasm varint implementation. However, wasm bytecode is almost always going to be delivered compressed with gzi

I think this is definitely an improvement over the wasm varint implementation. However, wasm bytecode is almost always going to be delivered compressed with gzip or brotli, so measurements of compression and speed should be taken after those. In particular, I'm wondering if a plain non-variable integer encoding would be best, considering how brotli and gzip operate on byte sequences.