[regression] isinstance complains about parametrized generics with any use of TypeAlias

[wesleywright]

Bug Report

Any attempt to call isinstance with a variable annotated as a TypeAlias seems to result in the error message error: Parameterized generics cannot be used with class or instance checks, even if the aliased type is non-generic. mypy_primer points at #18173 as the breaking commit, but that PR seems to have been intended to deal with type aliases of generic types; it seems that we're treating all TypeAlias-based aliases as generic types, which AFAICT is broader than intended.

To Reproduce

MyPy playground link, using current master

from typing import Any, TypeAlias

class Foo:
    pass

Alias: TypeAlias = Foo

def is_foo(x: Any) -> bool:
    return isinstance(x, Alias)

Expected Behavior

MyPy finds no errors, matching the runtime behavior. Pyright and mypy 1.4 both find no errors in my minimal example.

Actual Behavior

Per the playground link:

main.py:9: error: Parameterized generics cannot be used with class or instance checks  [misc]
Found 1 error in 1 file (checked 1 source file)

Your Environment

I found this when dogfooding mypy master commit c4e2eb7 against Dropbox's internal Python repo, but mypy_primer also reproduces the same thing, as does mypy playground.

[hauntsaninja]

cc @brianschubert in case you have time to take a look!

We could do e.g. the following to unblock

diff --git a/mypy/semanal.py b/mypy/semanal.py
index 034d8fb28..a92227d79 100644
--- a/mypy/semanal.py
+++ b/mypy/semanal.py
@@ -4022,7 +4022,7 @@ class SemanticAnalyzer(
             and not res.args
             and not empty_tuple_index
             and not pep_695
-            and not pep_613
+            # and not pep_613
         )
         if isinstance(res, ProperType) and isinstance(res, Instance):
             if not validate_instance(res, self.fail, empty_tuple_index):

Would have to disable some of the new test cases, but the spec is maybe a little aggressive in disallowing that type alias