Array creation

[jfmengels]

(Builds upon the fix-test-suite branch from #63, only the last commit is important here)

I made a transformer that removes the performance penalty for building Elm Array using something like Array.fromList [ 1, 2, 3 ], by precomputing the resulting the shape of the transformations result. This optimization only applies when encountering list literals. I did this by copying the compiled code that does this transformation inside this repository.

I made 2 benchmarks in JS-land to test the performance:

• https://jsbench.me/w6kwdopkj6/1 (10 items): Result is about 108x faster.

• https://jsbench.me/zrkwdnkgos (100) items: Result is about 2.5 times faster.
  (Note that the order of the test suite is not the same as above, sorry about the confusion)

I do suspect that the benchmark is not representative as it might be over-learning/over-optimizing this short piece of code, but I don't know how to better test this.

I don't know how to get this proposal further than this. I can't run the benchmarks locally (they fail for some reason). I should be able to add this transformer to the configuration that enables/disables transformations, and I'll do that soon, but that's as far as I know how to go without further help.

How do I go about validating that this improvement works out well in practice? And what are the missing steps to get this in?

[jfmengels]

Probably worth mentioning that this idea came from @robinheghan's post https://dev.to/skinney/de-throning-the-list-part-boron-185
I only came up with the idea of doing this through elm-optimize-level-2, most credit goes to Robin 😊