feat(biome_css_analyze): add rule useSortedProperties

[simon-paris]

Summary

Related: #3 (comment), #3167

This implements CSS declaration and rule sorting from the stylelint-order plugin. It sorts properties (and nested rules) in a consistent order, e.g:

  ! Properties should be sorted: display should be before color.
  
  > 1 │ .class { color: red; display: block; }
       │       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  
  i Safe fix: Sort these properties
  
    1    │ - .class·{·color:·red;·display:·block;·}
       1 │ + .class·{·display:·block;·color:·red;·}

The property order comes from the most popular preset config: stylelint-config-recess-order.

This is the equivalent stylelint config:

{
	extends: ['stylelint-config-recess-order'],
	rules: {
		order/order: [
		        "custom-properties",
		        "declarations",
		        "rules",
		        "at-rules"
	        ]
	},
}

The sort order is this: (ties stay in the same relative order)

1. Node kind:
   a. Custom properties
   b. The "composes" property
   c. Properties
   e. Nested rules and at-rules
   f. Other
2. Property name:
   a. By the order defined in PROPERTY_ORDER
3. Vendor prefix:
   a. By the order defined in VENDOR_PREFIXES
   b. Properties with no vendor prefix

Some possible issues:

• This is the only CSS rule that has a quickfix. I'm not sure if quickfixing CSS is stable. Seems to work though.
• stylelint-config-recess-order has an open PR that changes the order a little bit. It'd be best to wait for that to merge and sync the ordering before promoting this rule.
• stylelint-config-recess-order doesn't list every property. There are 230 (rarely used) properties that have no defined order. I'd rather not unilaterally decide where they go so I've left them out for now. This rule silences itself if any property exists that isn't in the list. The stylelint version doesn't raise errors for unknown properties but applying any autofix will move them to the end.
• I've had to duplicate some logic from the noShorthandPropertyOverrides rule because this rule would apply incorrect quickfixes in that case. I've made this rule silence itself if that rule would have fired. Not sure if that's the right approach. The stylelint plugin doesn't do this and applies the fix even when its unsafe.
• I've used LazyLock<HashMap> to implement some lookup tables for this rule. Not sure if that's a good method, I'm new to rust.

Test Plan

Unit tests are included.

[codspeed-hq]

CodSpeed Performance Report

Merging #4063 will degrade performances by 18.96%

_{Comparing simon-paris:feat/use-sorted-properties (379309b) with main (4007147)}

Summary

❌ 4 regressions
✅ 91 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	simon-paris:feat/use-sorted-properties	Change
❌	css_analyzer[bootstrap_18416142857265205439.css]	219.5 ms	240 ms	-8.55%
❌	css_analyzer[bulma_5641719244145477318.css]	624.5 ms	721.5 ms	-13.45%
❌	css_analyzer[foundation_11602414662825430680.css]	149.8 ms	184.8 ms	-18.96%
❌	css_analyzer[tachyons_11778168428173736564.css]	122.2 ms	143.7 ms	-14.96%

[ematipico]

Thank you @simon-paris for the contribution. I will look into this rule this week. As a very early feedback, this rule should be made an assist (apologies for the lack of documentation, I will try to fix that ASAP).

Here's an example of assist: https://github.com/biomejs/biome/blob/main/crates/biome_json_analyze/src/assists/source/use_sorted_keys.rs

[ematipico]

Sorry @simon-paris for the late, very late review.

I finally managed to understand your code and look at it. There's a lot of room for improvement. I suggest to check how use_sorted_keys is implemented, and emulate the same behaviour/implementation.

[simon-paris]

Thanks for the feedback!

I've addressed most of the issues already, I'll finish it up on monday.

[simon-paris]

Alrighty everything is addressed, ready for re-review.

Main changes:

1. Changed sorting to use a btree, structured code similar to use_sorted_keys
2. Replaced the complicated diagnostic message with a simple one
3. Removed all string allocations

[ematipico]

Second review. Main points:

• Technical documentation is not enough; we should strive to set in stone as much doc as possible
• User documentation is not enough; we should provide more exhaustive examples
• The diagnostic should be slightly improved to match our rule pillars
• The code, generally, can be simplified

[simon-paris]

@ematipico Thanks! I've addressed most of the comments, except the biome_string_case issue and the original_position issue which need clarification.

Main changes:

• Added better documentation and comments
• Added note to diagnostic, fixed ActionCategory parameter
• Use let else to reduce code nesting in several places

Thank you for your patience 🙏

[ematipico]

Yeah let/else is very powerful and useful. It's a new syntax that came out not long ago, so we recently started using it.