[ty] Fix panic when trying to pull types for subscript expressions inside Callable type expressions

[AlexWaygood]

Summary

We weren't storing types for subscript expressions inside the parameter list of Callable type expressions; this was the cause of the corpus test failures for two files in our vendored copy of typeshed.

Test Plan

• Removed the remaining typeshed entries from the KNOWN_FAILURES corpus list
• Added a standalone corpus test so that we still have coverage for this in case typeshed changes in the future
• Added an mdtest to cover the regression surfaced by primer on the first version of this PR

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

[AlexWaygood]

The new primer hits are because for this snippet:

from basedtyping import P

reveal_type(P)

We infer the type of P as ParamSpec | Unknown rather than ParamSpec, which means that we don't see it as a valid first argument to Callable. This is a pre-existing issue (and something of an edge case), we just see it in more cases now because we're now recursing inside Concatenate slices when inferring types.

[AlexWaygood]

With this PR, we also succeed in pulling types for the entirety of typeshed's stubs/ directory as well as its stdlib/ directory! Except for any files where the final segment is types.pyi: we still overflow our stack on all of those.

[dhruvmanila]

Looks good, thanks!

[dhruvmanila]

This is only being done for tuple expressions because otherwise it would already be present via the self.infer_type_expression(argument) call?

[AlexWaygood]

Yup, exactly!