[ty] Infer generic protocols from class objects (#27812)
## Summary
Passing a class object to a generic protocol parameter could leave the
protocol's type arguments as `Unknown`, even when its attributes or
methods supplied enough information to infer them. Ordinary instance
arguments and unions already used structural constraints, but individual
class objects fell through to inference that only considered `__call__`.
Use the existing structural constraint machinery for class literals,
specialized generic classes, and `type[C]` arguments. This preserves the
class object's own interface and also fixes `list(EnumClass)` inferring
`list[Unknown]`.
Fixes https://github.com/astral-sh/ty/issues/4291
## Test plan
Added mdtests for inference through plain class attributes, class
methods, static methods, ordinary-method protocols, specialized generic
class objects, and unions of class types. Callback regressions cover
preserving `T` when passing `type[T]` and inferring from a data member
alongside `__call__`. The tests also retain instance behavior and
rejection of an incompatible return type. Updated the enum iteration
test to expect the precise member type.
### Ecosystem
The ecosystem report includes a new `invalid-assignment` diagnostic in
[Scrapy's telnet
test](https://github.com/scrapy/scrapy/blob/488a762d7173922190ea027e0fb6f6d9d3b59880/tests/test_extension_telnet.py#L46),
which replaces a zero-argument bound method with `dict`. This PR
correctly infers `TelnetConsole` where the result was previously
`Unknown`, revealing an orthogonal, pre-existing limitation in ty's
instance-method assignment checking. The directly typed assignment is
rejected on both the base and PR revisions; the inference change does
not introduce that restriction. Compatible class-level method
replacement was addressed in
[#26158](https://github.com/astral-sh/ruff/pull/26158), while
instance-method replacement remains separate work related to
[ty#350](https://github.com/astral-sh/ty/issues/350). This is just
better type inference exposing an unrelated issue, so we accept it for
this PR.
Other ecosystem changes are desirable improved inference.
