Mypy reports invalid type when a value is a parameterized generic

Bug Report

When I attempt to use a parameterized generic as a value to build a signature at runtime, mypy complains about the parameter as being invalid, when the parameter comes from a variable.

To Reproduce

from inspect import Parameter

def signature_parameter(name: str, item_type: type) -> Parameter:
    return Parameter(name=name, kind=Parameter.KEYWORD_ONLY, annotation=list[item_type])

Expected Behavior

I would expect no error in this case (when the parameterized type is used as a value).

I know a variable cannot be used as a type parameter in a typing context (i.e. when used as a type annotation).
I understand that the following code would be wrong because variables cannot be used in type specifications:

var: type = int
lst: list[var] = [1, 2, 3]  # this is wrong

But in my case above, list[item_type] is not used as a type specification, but as a value, that is passed to the runtime to build a signature. I do not think mypy should flag such use of values. It should simply consider the value to be of type types.GenericAlias.

Actual Behavior

typing-8.py:6: error: Variable "item_type" is not valid as a type  [valid-type]
typing-8.py:6: note: See https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
Found 1 error in 1 file (checked 1 source file)

Your Environment

• Mypy version used: mypy 1.12.0+dev.6a0657e5959ba1777c4d427f8f355d499035d145 (compiled: no)
• Mypy command-line flags: None
• Mypy configuration options from mypy.ini (and other config files):

[tool.mypy]
enable_incomplete_feature = "NewGenericSyntax"
disable_bytearray_promotion = true
disable_memoryview_promotion = true
check_untyped_defs = true
warn_unreachable = true
warn_redundant_casts = true
warn_unused_ignores = true

• Python version used: 3.12

Comienza con el reproductor typing-8.py y ejecuta mypy bajo Python 3.12 para confirmar el diagnóstico [valid-type] para list[item_type]. Sigue la ruta de comprobación de tipos responsable de ese diagnóstico y añade una prueba de regresión que muestre que este valor de anotación en tiempo de ejecución se acepta; el trabajo estará terminado cuando la prueba pase sin suprimir el error.