perf(es/ast): Use Atom in some places

[kdy1]

Description:

As this is tradeoff of comparison/hash time and global lock, I decided to use Atom for only places I can be sure.

Related issue (if exists):

[kdy1]

@kwonoj As it's a type, how long is too long does not exist.
It's used unconditionally, and the important point is not likely to be duplicated, not length

[overlookmotel]

Argh! I'm too late to the party. You move so damn fast!

I had an idea about how it might be possible to combine the advantages of JsWord and the new Atom in a single entity.

In short: have a shared-across-threads static set (read only, so no mutex required) pre-populated with common words + a per-thread dynamic set (for words not in the static set). The static set could use perfect hashes, and the dynamic set not. Both could use the faster FxHasher.

However, work is completely crazy today so it'll have to wait until weekend for me to outline this idea further. I'll open an issue as soon as I have time.

[kdy1]

I merged it after verifying the performance of other operations, but yeah, I only used Atom for places where we don't need perfect hashing

[overlookmotel]

Quick thought: Since the Atoms are never shared across threads, couldn't they use Rc instead of Arc? My understanding is Rc can have less overhead.

[kdy1]

Nope. they are moved across threads