Add opt-in flag for method hook defining-class fullnames

[itxsamad1]

Summary

This adds an opt-in flag for the defining-class behavior discussed in #19181.

By default, plugin method hooks still receive the call-site class name (e.g. Derived.method), so existing plugins keep working. If you set use_method_hook_defining_class = True in config (or pass --use-method-hook-defining-class), hooks get the class where the method was defined instead (Base.method).

Fixes #19181

Test plan

• Ran the plugin hook tests locally, including the new defining-class regression test with the flag enabled

[A5rocks]

I think we should fix this instead of documenting errata, and that seems to be consensus in the issue too.

[itxsamad1]

Updated per @A5rocks feedback — this is now a code fix, not a docstring errata update.

What changed: ExpressionChecker.method_fullname() now uses TypeInfo.get_containing_type_info() to pass the defining class name (e.g. Base.method) to get_method_signature_hook / get_method_hook, instead of the call-site class (Derived.method).

Tests: Added testMethodSignatureHookUsesDefiningClass with a plugin that hooks only on __main__.Base.method and verifies Derived().method() triggers it. Updated the arg_names test plugin accordingly for inherited methods.

Note: plugins that registered hooks on derived-class fullnames only (without also hooking the base class) may need updating — this aligns with the original documented intent per the issue discussion.

[itxsamad1]

Updated per @A5rocks feedback — this is now a code fix, not docs-only.

method_fullname() uses TypeInfo.get_containing_type_info() so inherited calls like Derived().method() invoke plugin hooks with Base.method (the defining class), matching the documented/intended semantics discussed in #19181.

Added regression test estMethodSignatureHookUsesDefiningClass and ran 276 plugin-related tests locally.

[A5rocks]

Sorry but the issue also describes a backward compatibility approach to take. Anyways if this PR is purely LLM generated then we will close it.

[itxsamad1]

Sorry about the duplicate comments earlier — that was my mistake.

Re the backward-compat point: I reworked this so the old behavior stays the default. Hooks still get the call-site class unless you opt in with \use_method_hook_defining_class = True\ (or --use-method-hook-defining-class).

The new test only enables that flag, so existing plugin tests should behave the same as before. Happy to adjust naming or rollout if you'd prefer a different migration path.

[github-actions]

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