Missing `TypeIs` narrowing for variadic tuples

I noticed this behavior when investigating a similar issue in pyright.

Mypy isn't correctly narrowing tuples in some cases.

from typing_extensions import TypeIs


def is_tuple_of_strings(v: tuple[int | str, ...]) -> TypeIs[tuple[str, ...]]:
    return all(isinstance(x, str) for x in v)


def test1(t: tuple[int]) -> None:
    if is_tuple_of_strings(t):
        reveal_type(t)  # Should be Never ✅ 
    else:
        reveal_type(t)  # Should be tuple[int] ✅ 

def test2(t: tuple[str, int]) -> None:
    if is_tuple_of_strings(t):
        reveal_type(t)  # Should be Never ✅ 
    else:
        reveal_type(t)  # Should be tuple[str, int] ✅ 

def test3(t: tuple[int | str]) -> None:
    if is_tuple_of_strings(t):
        reveal_type(t)  # Should be tuple[str] ✅ 
    else:
        reveal_type(t)  # Should be tuple[int] or tuple[int | str] ❌ (mypy: Never)

def test4(t: tuple[int | str, int | str]) -> None:
    if is_tuple_of_strings(t):
        reveal_type(t)  # Should be tuple[str, str] ✅ 
    else:
        reveal_type(t)  # Should be tuple[int | str, int | str] or tuple[int, int | str] | tuple[str, int] ❌ (mypy: Never)

def test5(t: tuple[int | str, ...]) -> None:
    if is_tuple_of_strings(t):
        reveal_type(t)  # Should be tuple[str, ...] ✅ 
    else:
        reveal_type(t)  # Should be tuple[int | str, ...] ❌ (mypy: Never)

def test6(t: tuple[str, *tuple[int | str, ...], str]) -> None:
    if is_tuple_of_strings(t):
        reveal_type(t)  # Should be tuple[str, *tuple[str, ...], str] ❌ (mypy: tuple[str, Never, str])
    else:
        reveal_type(t)  # Should be tuple[str, *tuple[int | str, ...], str] ❌ (mypy: Never)

Empieza reproduciendo los seis ejemplos con mypy y comparando cada resultado de reveal_type con el estrechamiento esperado en el issue. Rastrea la implementación del estrechamiento de tipos para TypeIs y las tuplas variádicas, y añade después cobertura de regresión para estos casos. La tarea estará terminada cuando los tipos revelados coincidan con los resultados esperados sin regresiones.