Use Object.hash to hash raw bytes

[jakemac53]

Avoids decoding values during hashing just to re-encode them as bytes to hash, when stored as raw bytes. Ultimately this mostly just applies to Strings, but that is still impactful. We also avoid a lot of type checks, instead doing switches based on the known or encoded types.

For now I am just using Object.hash and Object.hashAll but we could explore alternatives later on as well, although it would mean passing a byte sink around.

This drops the total time for the large benchmark another 10% ish on my machine.

[davidmorgan]

A couple of high level notes:

• Needs some test coverage :)
• Pretty sure Object.hash is not what we want in the end, 32 bit Jenkins hash is pretty small; doesn't need changing now though

[jakemac53]

• Needs some test coverage :)

Huh, I definitely wrote a test at some point previously covering this but I can't find it... 🤔 . I think I must have never landed it lol.

[jakemac53]

Aaaand now I am on an adventure fixing growable maps 🤣 (apparently if they have growable maps as values they stack overflow 😄 , or something like that). turns out I was creating cyclic structures on accident

[jakemac53]

• Pretty sure Object.hash is not what we want in the end, 32 bit Jenkins hash is pretty small; doesn't need changing now though

Yeah it is pretty explicitly documented as "collisions are OK, just might make maps slow" 🤣 . So, we very likely do want to do something else here.