Optimization and benchmarking

[pbiggar]

I was doing some benchmarking and discovered that running fuzzbuzz in Dark was taking a ridiculously long time, like 800ms. It was only taking 70ms in OCaml (including the http and serialization overhead, which was probably most of it).

The problem was that pure functions were all being marked as impure, and so every function call was having its arguments and result saved.

This adds the benchmarking, the fix and also some other major changes:

• removes TaskOrValue and switches the Ply (which is a ValueTask), which is roughly the same. Ply is surprisingly slightly slower, but it's in the noise:
  • taskv -> uply
  • value -> Ply.Ply
  • TaskOrValue<a'> -> Ply.Ply<'a>
• There were a few sequential async List and Map functions, those are now in the Task namespace and also in the Ply namespace (Ply.List.iterSequentially for example). This replaces uses of map_s and similar.
• removes a couple of unnecessary taskv from interpreter
• note a few places where we could do some optimization
• fix an error when were would only load traces from impure functions
• fix List.sortByComparator which wrapped a uply in a try (instead of the other way around). Also added a note to the coding guide about this.