Better datetime/date string parsing performance

[texodus]

Tl;DR 7x performance improvement parsing non-ISO-8601 datetime strings in CSV or JSON formats.

I've been analyzing Perspective performance using this NYC Open Data 500k row CSV file, which loads fine but takes ~22 seconds on my computer to parse:

Looking into the profile analysis shows a lot of long calls out from WASM into JS to call strptime, which Perspective uses for anything not ISO 8601 formatted date or datetime strings. In Emscripten, strptime is implemented as a JavaScript foreign call which turns out to be quite expensive, as well as requiring unnecessary string copying.

Perspective ultimately callsstrptime through Arrow, which is used internally for CSV and any other string-to-datetime parsing (including in JSON formats). It turns out though, Arrow itself has a vendored C++ implementation of strptime which it uses for Windows builds. Patching Arrow to use its own vendored strptime implementation for Emscripten as well reduces the original runtime to ~3s:

This PR only applies this fix for Emscripten (WebAssembly) builds. I haven't yet tested Python, but I don't expect this fix to be applicable for Python as these implementations are likely identical in this context.