feat(organize_import): utilities for ordering import sources

[Conaclos]

Summary

Implements #4239.

In the RFC (#3015), I proposed to parse import sources in order to compare them in a structural way.
At the time of writing the RFC it took a lot of time trying to decide which structure to parse.
I left the RFC a bit loose because I couldn't find a satisfactory structure.

I fell into the same problem again. I am not 100% happy with the result and I am open to suggestions.
However, the current implementation should be good enough, and we can improve it later.

The implementation is in biome_js_analyze/src/assists/source/organize_imports/util.rs and biome_string_case/src/lib.rs.

In contrast to the spec, I decided to avoid a parsing phase (except for paths).
Instead, we just look at the first bytes of an import source to determine its kind.
An import source can be one of the following kind:

• Unknown: all import paths without a recognizable kind
• Uri, e.g. https://example.org
• ProtocolPackage, e.g. node:test, npm:@scope/package
• Package, e.g. package
• Alias: Node.js subpath imports and TypeScript aliases.
  This corresponds to import sources that start with @/, #, ~, or %
  /// Import sources that start with /, ./, or ../
• Path: import sources that start with /, ./, or ../

Import paths are first ordered according to their kind (order of kind is reflected by their position in the previous list).
Note that unrecognized import source kinds are placed first.
I am unsure if we should keep or remove Unknown and choose a kind instead (probably Alias).

Two import sources with identical kind are then ordered by a custom order:

• Order between paths follows the proximity order (../../ < ../ < ./)
• Other kinds use a string collation (ImportSourceAsciiCollator) tailored for paths and URIs.
  The collation is case-insensitive (A < a < B < b), compares numbers in a human-friendly way (9 < 10).
  Special characters such as / and @ are carefully ordered to yield a more natural ordering for import sources.

I added the base of the collation algorithm in biome_string_case because it is kind of related to character case.
Another option is to create a dedicated crate biome_string_collator or a crate biome_str_utils.

Note: This PR can be merged into main as the utilities introduced are not yet in use.

Alternative design

Instead of creating a custom collation, we could use the conventional Unicode Collation reduced to the ASCII set (biome_string_case::CldrAsciiCollator) and to parse import sources in a more advanced structure. I started with this design in mind and find it a bit too complex. However, I am still open to revisiting it if you think it could be better.

Test Plan

I added tests in the source code.

[github-actions]

Parser conformance results on

js/262

Test result	main count	This PR count	Difference
Total	48420	48420	0
Passed	47212	47212	0
Failed	1208	1208	0
Panics	0	0	0
Coverage	97.51%	97.51%	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	6578	6578	0
Passed	2204	2204	0
Failed	4374	4374	0
Panics	0	0	0
Coverage	33.51%	33.51%	0.00%

ts/babel

Test result	main count	This PR count	Difference
Total	680	680	0
Passed	607	607	0
Failed	73	73	0
Panics	0	0	0
Coverage	89.26%	89.26%	0.00%

ts/microsoft

Test result	main count	This PR count	Difference
Total	18489	18489	0
Passed	14154	14154	0
Failed	4335	4335	0
Panics	0	0	0
Coverage	76.55%	76.55%	0.00%

[codspeed-hq]

CodSpeed Performance Report

Merging #4313 will not alter performance

_{Comparing conaclos/import-sorter-revamping-source-order (8cafba8) with main (30a016b)}

Summary

✅ 101 untouched benchmarks