Reject type variable whose upper bound is Self

[EmmanuelNiyonshuti]

Fixes #21960

First attempt by an ai agent, See: #21961

[EmmanuelNiyonshuti]

I’m not entirely sure if this is the kind of validity that this comment is referring to.

[wolfgang-aura]

I filed #21961 for this issue with the narrower Self-location framing. Yours
is the right shape. I built the same check independently before I saw this PR
and it agrees line for line, so I am closing mine. Two things from running it.

The check covers more than Self, and that extra coverage is untested here. On
the base commit both of these are silently accepted, and with your patch both
error:

class D[U]:
    def g[T: U](self, x: T) -> None: ...

class F:
    def nested[T: list[Self]](self, x: T) -> None: ...

The first is the plain "bound parametrized by a type variable" case the spec
forbids, with no Self in it. A test case for it would pin the rule rather
than the symptom.

On fallout: with the check applied to ae39cdb, pytest mypy/test/testcheck.py was 8241 passed, 15 skipped, 7 xfailed, so no existing
test relies on a generic bound.

Minor: TYPE_VAR_GENERIC_BOUND_TYPE reads "can not be parametrized", while its
neighbour TYPE_VAR_GENERIC_CONSTRAINT_TYPE reads "cannot be parametrized".
Worth matching.

[github-actions]

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