Unpack a tuple subclass with an unbounded part like a plain tuple

Summary:
Unpacking a `tuple` subclass whose tuple has an unbounded part, such as SQLAlchemy's `Row`, gave every target the union of all the element types. `iterate` only treated a subclass as a tuple when its tuple was fixed-length. Any other subclass fell through to the iterable protocol, which knows only the element type of `__iter__`, so `a, b, c = t` with `t: Variadic[str, int | None, *tuple[Any, ...]]` made `a` and `b` `int | str | Any | None`.

Now a subclass whose tuple has fixed elements before or after the unbounded part is iterated as that plain tuple, which already keeps each fixed position's type. So `a` is `str`, `b` is `int | None`, `c` is `Any`, and a starred `*rest` is `list[Any]`, matching how a plain tuple with the same types unpacks.

Fixed-length subclasses keep their existing path. Subclasses whose tuple has no fixed elements also keep the iterable protocol: routing them through the plain-tuple path added a false `bad-unpacking` error on a subclass of a generic tuple subclass (`class Nested[*Ts](Variadic[*Ts])`), which unpacks fine at runtime. That case still loses position types and is a separate gap.

Fixes https://github.com/facebook/pyrefly/issues/5286

Reviewed By: samwgoldman

Differential Revision: D127237256

fbshipit-source-id: 9faba386fe7a15ebe324261ef3af37132db3d060