Reject Self in type parameter bounds

[Madin-H23]

Fixes #21960

Rejects Self used as the bound of a type parameter, in both PEP 695 syntax and legacy TypeVar(bound=...) declarations, by reusing the existing prohibit_self_type mechanism in semantic analysis:

• analyze_type_param now analyzes a PEP 695 upper bound with prohibit_self_type="a type variable bound". Class-level bounds previously hit the existing "only allowed in annotations within class definition" check; they now get the more precise new message.
• The legacy bound= path passes the same context through get_typevarlike_argument and expr_to_analyzed_type (new pass-through parameters), so TypeVar("T", bound=Self) inside a class body is also rejected. Module-level use was already rejected.

Notes:

• The legacy path reports the follow-up "TypeVar bound must be a type" alongside the new error. I kept both for now; happy to fine-tune if a single error is preferred.
• Constraints (TypeVar("T", Self, int)) and PEP 695 defaults are left unchanged — out of scope for this issue, can follow up if maintainers want those rejected too.

New cases are in testTypingSelfInvalidAsTypeParameterBound (check-selftype.test), covering in-class method-level and class-level PEP 695 bounds plus the legacy form.

[github-actions]

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

[Madin-H23]

Closing in favor of #21967, which predates this PR by three weeks and is from the issue reporter. My apologies for the duplicate — I located it only after opening this one; searching by the issue's keywords missed it because the PR title is phrased differently. I'm leaving a couple of implementation notes from my own pass on #21967 instead.