[ty] Optimize protocol subtyping by removing expensive and unnecessary equivalence check from the top of Type::has_relation_to()

[AlexWaygood]

Summary

A full equivalence check at the top of Type::has_relation_to() is unnecessary. All our mdtests, and all our property tests, pass if it is replaced with a much cheaper naive equality check. The reason for this is that equivalence is generally the same as equality for atomic types (types that cannot contain other types); and "non-atomic" types (types that can contain other types within them) all have comprehensive subtyping logic lower down in the Type::has_relation_to() method. For every non-atomic type in our model currently, our subtyping logic is sufficiently generalized to also cover equivalent types; there's no need to check for equivalence separately.

Up till now, the inefficiency here hasn't mattered that much, because comparing two nominal types for equivalence is generally pretty cheap. (It is not cheap for large unions, large intersections or large tuples (or combinations of the three), but these are all somewhat rare.) Unfortunately, however, comparing two structural types for equivalence can be extremely expensive: determining whether two protocols are equivalent can necessitate a full structural check between the two underlying interfaces. That means that this PR yields a 7% speedup when checking DateType, which has some large protocols in it. We also see 1% speedups on a lot of other micro- and macro-benchmarks for ty, and no slowdowns: https://codspeed.io/astral-sh/ruff/branches/alex%2Fremove-equivalent-check?runnerMode=Instrumentation.

Test plan

• Existing tests all pass
• QUICKCHECK_TESTS=100000 cargo test --release -p ty_python_semantic -- --ignored types::property_tests::stable passes
• Clean mypy_primer report

[github-actions]

mypy_primer results

No ecosystem changes detected ✅
No memory usage changes detected ✅

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #19230 will improve performances by 7.33%

_{Comparing alex/remove-equivalent-check (6c9d9fb) with main (35a33f0)}

Summary

⚡ 1 improvements
✅ 39 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	DateType	263.6 ms	245.6 ms	+7.33%

[sharkdp]

Very nice! Thank you