Introduce benchmarks and tracking of performance over time

[Janiczek]

We'd like to have some performance tracking over time, similar to https://arewefastyet.com/

Probably don't bother with setting up a webpage that will hold the historic data and add to them - as a first iteration, let's just have a script (possibly in a different repo) that:

• runs the benchmarks suite on all commits in this repo
• outputs values for each commit (those can be N/A if that particular revision didn't support that Elm input and it errors out - that's to be expected)
• and then generates some graphical representation (plaintext generation of SVG? D3? terezka/line-charts / elm-plot / ...? Feel free to choose the easiest solution) or at least a table with the numbers.

We'll need:

• [ ] a corpus of interesting projects / modules / expressions to run the benchmarks on
• [ ] the ways to run elm-in-elm on each input - possibly this:
  • [ ] parse
  • [ ] parse + desugar
  • [ ] parse + desugar + typecheck
  • [ ] parse + desugar + typecheck + optimize
  • [ ] parse + desugar + typecheck + optimize + emit
  • maybe with some numerical trickery we can separate those cases from each other so that we know "the desugar phase took this long"
• [ ] a script to run benchmarks on each of these
• [ ] a script to convert the raw numbers to a graphical representation

The output files should also have some metadata about the machine they were run on (Node version, OS, CPU, memory, etc?)

We can hook you up with a new repository (say elm-in-elm/benchmarks) when needed.