refactor(config): reduce size by 350% using exact rule options

[Conaclos]

Summary

Part of #1263.
This should also help for #1795.

This PR reduces by 352% (from 4016 bytes to 1140 bytes) the memory usage of biome.json.
Most of the memory usage comes from the RuleConfiguration type (16 bytes) that is present for every rule.
RuleConfiguration is an enum with two variants, thus it took the size of its biggest variant: RuleWithOptions.
Most of the rules have no options, so this is just a waste of memory.

I managed to reduce the size of RuleWithOptions to only 1 byte for a rule without options.
This reduces the size of RuleConfiguration to 2 bytes :)
To do that, I added a type parameter to RuleConfiguration and RuleWithOptions.
This type parameter is set to the Options type of the rule. Most of the rule has Rule::Options set to ().
Rules with options took a variable amount of memory depending on the memory layout of their options.
I used the same trick as my previous PR (#1283) to ensure a maximum usage of 16 bytes: I wrapped the Options type in a Box when its size is larger than 16 (this is the case for 3 rules).

A nice side effect is that we get exact JSON schema.

We now generate an options.rs file in biome_js_analyze and biome_json_analyze that exports the options for the rules.
I had to make the rules public to satisfy the Rust compiler: it reported the use of a crate-scoped type in the public interface. Unfortunately disabling the Clippy check is not enough: the compiler still complained.

I also had to fix a bug in our deserializable derive macro that handled wrongly constrained generics.

Test Plan

CI should pass.

[netlify]

✅ Deploy Preview for biomejs canceled.

Name	Link
🔨 Latest commit	845a9d8
🔍 Latest deploy log	https://app.netlify.com/sites/biomejs/deploys/65d625b594df6e0008db139f

[github-actions]

Parser conformance results on

js/262

Test result	main count	This PR count	Difference
Total	49701	49701	0
Passed	48721	48721	0
Failed	980	980	0
Panics	0	0	0
Coverage	98.03%	98.03%	0.00%

jsx/babel

Test result	main count	This PR count	Difference
Total	40	40	0
Passed	37	37	0
Failed	3	3	0
Panics	0	0	0
Coverage	92.50%	92.50%	0.00%

symbols/microsoft

Test result	main count	This PR count	Difference
Total	6322	6322	0
Passed	2036	2036	0
Failed	4286	4286	0
Panics	0	0	0
Coverage	32.20%	32.20%	0.00%

ts/babel

Test result	main count	This PR count	Difference
Total	662	662	0
Passed	593	593	0
Failed	69	69	0
Panics	0	0	0
Coverage	89.58%	89.58%	0.00%

ts/microsoft

Test result	main count	This PR count	Difference
Total	17646	17646	0
Passed	13439	13439	0
Failed	4205	4205	0
Panics	2	2	0
Coverage	76.16%	76.16%	0.00%

[codspeed-hq]

CodSpeed Performance Report

Merging #1876 will not alter performance

_{Comparing conaclos/config/exact-options-type (845a9d8) with main (69f9031)}

Summary

✅ 93 untouched benchmarks

[arendjr]

Cool stuff!

[Conaclos]

I guess a trick like TypeId::of::() == TypeId::of::<()>() might work?

Good suggestion! Unfortunately this requires a 'static lifetime on every generic. I managed to avoid that by using std::mem::size_of_val(x) == 0 instead.

However, one snapshot (correct_root) is failing with an internal error (data did not match any variant of untagged enum RuleConfiguration). It seems that skipping the serialization of options affects in some way the desexualization.

I think it is fine to keep the current caveat because it is unlikely that someone uses level without options. Thus, serializing to { level: _, options: null } is unlikely.