Resolve bare Protocol and Generic references in type position

[istoolsfox]

Follow-up to the typeshed declaration changes: Protocol is now declared as type[_Protocol] (python/typeshed#16386), the same shape Generic has had since python/typeshed#14583. Both mirror the runtime, where the two names are classes on 3.12+.

Mypy already honors those declarations in expression context -- reveal_type(Protocol) gives type[typing._Protocol] with the synced stubs -- but a bare reference in type position still failed:

from typing import Protocol

x: type[Protocol] = Protocol
# error: Variable "typing.Protocol" is not valid as a type  [valid-type]

This resolves the bare reference to Instance(_Protocol) / Instance(_Generic), so the example checks clean once the stub sync lands. Details:

• References with arguments (Protocol[T] in type position) keep failing as before -- that is class-base syntax, not a type. Empty subscripts (Protocol[()]) also keep failing.
• Class bases are untouched; semanal handles those separately.
• If the stubs in use do not declare the private classes, the regular "not valid as a type" error is kept instead of resolving to Any, so behavior changes only together with the stub declarations -- safe to merge in either order relative to the typeshed sync.
• The expression-context reproducer from typing.Protocol is explicitly treated as not a type #21940 is already fixed by the stub change itself; this closes the remaining type-position gap.

The unit-test fixtures declared Generic = 0 / Protocol = 0, which would silently change meaning under the new branch, so the typing fixtures now mirror the real declarations.

One divergence worth flagging: pyright 1.1.414 rejects these examples too (reportInvalidTypeForm), and TypeVar("T", bound=Protocol) gets a nearly-empty upper bound under this change. I went with "the typeshed declaration is the contract"; happy to adjust if bare references in type position should keep being rejected.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅