cachetools: Subclassing TTLCache[KT, VT] loses generic type inference on instantiation

Description

When subclassing TTLCache[KT, VT] with explicit generic parameters,
type checkers (tested with Pyright and mypy) fail to infer the correct
type on instantiation, reporting _SubClass[Any, Any, float] instead
of _SubClass[KT, VT].

Reproduction

from typing import Any
from cachetools import TTLCache

class _TrackedTTLCache[KT, VT](TTLCache[KT, VT]):
    def popitem(self) -> tuple[KT, VT]:
        key, value = super().popitem()
        return key, value

cache: _TrackedTTLCache[str, int] = _TrackedTTLCache(maxsize=128, ttl=600)
#     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
# Error: Expected type '_TrackedTTLCache[str, int]',
#        got '_TrackedTTLCache[Any, Any, float]' instead

Expected behavior

_TrackedTTLCache(maxsize=128, ttl=600) should be inferred as
_TrackedTTLCache[str, int] — consistent with how the explicit
annotation on the left-hand side is declared.

Root cause

The stub signature for TTLCache.__init__ likely includes float
as a third implicit type parameter (from the timer argument),
which leaks into subclass inference and overrides the declared
generic parameters.

Workaround

from typing import cast

cache = cast(
    "_TrackedTTLCache[str, int]",
    _TrackedTTLCache(maxsize=128, ttl=600),
)

Environment

• cachetools version: 7.1.1
• PyCharm linter
• Python: 3.12

Start by locating the typeshed stub for TTLCache.init and any related type-checker tests; inspect how the timer argument contributes to generic inference. Reproduce the subclass example with mypy and Pyright, then verify that instantiating _TrackedTTLCache(maxsize=128, ttl=600) is inferred as _TrackedTTLCache[str, int] without the cast workaround.