[ty] Use more parallelism when running corpus tests

[AlexWaygood]

Summary

Now that we've moved the corpus tests to the ty_python_semantic crate, running cargo test -p ty_python_semantic takes an annoying amount of time locally. (This also hasn't been helped by the fact that we run a lot more corpus tests after #18531.) We can speed things up quite a bit by splitting the corpus tests up more, so that they run in parallel.

Test Plan

On main locally:

~/dev/ruff (main)⚡ % cargo test -p ty_python_semantic --test corpus              
   Compiling ty_python_semantic v0.0.0 (/Users/alexw/dev/ruff/crates/ty_python_semantic)
    Finished `test` profile [unoptimized + debuginfo] target(s) in 0.81s
     Running tests/corpus.rs (target/debug/deps/corpus-1d987a322669ccb0)

running 6 tests
test corpus_no_panic ... ok
test linter_stubs_no_panic ... ok
test parser_no_panic ... ok
test linter_af_no_panic ... ok
test linter_gz_no_panic ... ok
test typeshed_no_panic ... ok

test result: ok. 6 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 8.63s

On this PR branch locally:

~/dev/ruff (alex/split-corpus-tests-more)⚡ % cargo test -p ty_python_semantic --test corpus
    Finished `test` profile [unoptimized + debuginfo] target(s) in 0.08s
     Running tests/corpus.rs (target/debug/deps/corpus-1d987a322669ccb0)

running 12 tests
test corpus_no_panic ... ok
test linter_no_panic::_a_e_expects ... ok
test linter_no_panic::_g_o_expects ... ok
test linter_stubs_no_panic ... ok
test parser_no_panic ... ok
test typeshed_no_panic::_f_k_expects ... ok
test typeshed_no_panic::_l_p_expects ... ok
test typeshed_no_panic::_a_e_expects ... ok
test linter_no_panic::_q_z_expects ... ok
test typeshed_no_panic::_q_z_expects ... ok
test linter_no_panic::_p_expects ... ok
test linter_no_panic::_f_expects ... ok

test result: ok. 12 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 3.77s

I.e., a speedup of around 5s

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

[MichaReiser]

This could make the tests on ci (or systems with fewer cores in general) somewhat slower because we now have to infer typeshed more often but I think this change still makes sense.