Implement core logic in WebAssembly for speed

[davidje13]

WebAssembly is now well supported and could improve the speed of the core algorithm.

Initially this can continue to be single-threaded (with the webassembly function invoked inside the existing Web Worker) but later it could be further improved by using pthreads.

A compiled C++ or Rust method seems best (Rust may be more appropriate since the existing tooling targets individual methods, whereas the emscripten C++ tooling targets whole applications)

There is no need to maintain fallback javascript logic, as the only browser which does not support WebAssembly is IE, and this is already not a supported browser.

• https://webassembly.org/getting-started/developers-guide/
• https://developer.mozilla.org/en-US/docs/WebAssembly
• https://developers.google.com/web/updates/2018/10/wasm-threads (future enhancement)
• https://developer.mozilla.org/en-US/docs/WebAssembly/Rust_to_wasm
• https://rustwasm.github.io/docs/book/reference/crates.html

[davidje13]

Rust proved undesirable as even simple functions produced ~300kb wasm files (with all possible optimisations enabled). Using C instead, with a custom binder, which has minimal bloat.

Moving a few small but core functions over has already produced a better-than 10% speedup, so this looks promising as an approach.

[davidje13]

Migrating generate has shown a substantial speedup, but a naïve migration of pow has caused a slowdown and limits the range to 12 months due to size limitations (branch wip-wasm-pow). An alternative data structure would be good for pow (e.g. hashmap or tree).

This involves a lot of shifting data from javascript to wasm and back. It might be better to share all wasm memory between threads and offer a large pool of data which all threads can write to and return pointers to. This will require some thought (and I don't know how the stack is managed with multi-threaded wasm)