perf(rome_analyze): optimize code actions diff generation

[leops]

Summary

The generation of code action diffs shows up a lot when profiling the analyzer, so I've optimized it a bit by making it skip over irrelevant subtrees. For find_diff_range this implies comparing the green nodes between the old and new version of the tree, so I've made the key method public on SyntaxNode / SyntaxToken / SyntaxElement

Test Plan

The snapshot tests for the analyzer still pass, as well as the unit tests for find_diff_range although I had to change the expected values since the function now compares the green nodes instead of the text content

[github-actions]

Playground for commit 32c92a0

[github-actions]

Parser conformance results on ubuntu-latest

js/262

Test result	main count	This PR count	Difference
Total	45878	45878	0
Passed	44938	44938	0
Failed	940	940	0
Panics	0	0	0
Coverage	97.95%	97.95%	0.00%

jsx/babel

Test result	main count	This PR count	Difference
Total	39	39	0
Passed	36	36	0
Failed	3	3	0
Panics	0	0	0
Coverage	92.31%	92.31%	0.00%

symbols/microsoft

Test result	main count	This PR count	Difference
Total	5946	5946	0
Passed	388	388	0
Failed	5558	5558	0
Panics	0	0	0
Coverage	6.53%	6.53%	0.00%

ts/babel

Test result	main count	This PR count	Difference
Total	588	588	0
Passed	519	519	0
Failed	69	69	0
Panics	0	0	0
Coverage	88.27%	88.27%	0.00%

ts/microsoft

Test result	main count	This PR count	Difference
Total	16257	16257	0
Passed	12393	12393	0
Failed	3864	3864	0
Panics	0	0	0
Coverage	76.23%	76.23%	0.00%

[ematipico]

!bench_analyzer

[ematipico]

The action died: https://github.com/rome/tools/runs/7249709125?check_suite_focus=true

[leops]

The action died: https://github.com/rome/tools/runs/7249709125?check_suite_focus=true

That's probably due to a timeout, until #2847 is merged the current benchmark suite for the analyzer is pretty much useless

[IWANABETHATGUY]

I was wondering any reason why we don't use a string-based diff algorithm?

[leops]

I was wondering any reason why we don't use a string-based diff algorithm?

This is technically all about converting a CST-based diff into a text-based diff, but for the initial version the diff only consist of a single range-replace operation since it was easier to just implement that first. We do have the capability to eventually represent more complex string diffs with the rome_text_edit crate, and the switch could happen sooner rather than later if the theoretically simpler algorithm we have now continues to require some tweaking to run correctly