Pre-filter and sort rules for 400% perf boost

[TrevorBurnham]

Closes #193

Currently, the Lint() function loops through all 4361 rules on every node. The cost of all that looping really adds up! For most projects, most of those rules are irrelevant because they only fail for very old or esoteric browsers. In the project I used to test this branch, the browser list is

[
  'last 3 chrome version',
  'last 3 firefox version',
  'last 2 edge version',
  'last 2 safari version',
  'IE 11'
]

and only 1679 of the rules potentially fail. For a project that only supports modern browsers (no IE11), the relevant set of rules would be much smaller.

This PR filters the set of rules based on the targeted browser list, removing the need to iterate through irrelevant rules on every node. It does that filtering only once per browser list. This change reduces the time it takes to run compat/compat against my test project from ~5.5s to ~2.5s. 🚀

---

I've marked this as WIP because the code here is very messy, so please take it as a proof of concept. If you like this change, let me know and I'll tidy it up!

[TrevorBurnham]

Now that #245 is merged, I've rebased this branch. I'll try tidying up the code later this week.

[TrevorBurnham]

I've added another major optimization here: Instead of iterating over all the rules on every node (checking that the rule type matches the node type on each iteration), I pre-sort the rules by their astNodeType. So now CallExpression nodes only run against CallExpression rules, NewExpression nodes only run against NewExpression rules, etc. I've also made each node type run only the validation logic required for that node type, instead of running a type-agnostic validation function.

The impact of these changes is huge. On the current master, my test project takes ~5.5s to run compat/compat. On this branch, it only takes ~1s! 🎉

The code could still use some cleanup, but I'm thrilled with the improvement. For comparison, this takes compat/compat from being by far the most expensive rule in my project's linter config to ranking in third place, between react/no-direct-mutation-state and react/sort-comp. 🥉

[TrevorBurnham]

@amilajack OK, I've done some refactoring to tidy up the code on this branch and improve Flow coverage. I hope it's reasonably clear and well-organized now. Please let me know if there are any changes you'd like to see!

[amilajack]

Awesome! This is a huge improvement to this project! Thank you so much 🎉. I'll add this to the 3.3.0 release as well.