[asukaminato0721]

Summary

Fixes #4256

BeforeValidator and PlainValidator constructor input inference.
Support across reusable TypeAlias annotations.
Correct right-to-left ordering for chained before validators, matching Pydantic's behavior (https://pydantic.dev/docs/validation/latest/concepts/validators/#ordering-of-validators).
Support for fields with defaults while preserving the stored field type.

Test Plan

add test

[meta-codesync]

This pull request has been imported. If you are a Meta employee, you can view this in D114356273. (Because this pull request was imported automatically, there will not be any future comments.)

[github-actions]

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

[rchen152]

Unfortunately, I think this has the same issue as #3798 - it works through converter_param, which also affects attribute assignment, when we should only modify __init__ parameter annotations. Some other notes:

• The pydantic test stubs don't have functional_validators.py, so the tests don't exercise the same code paths that the real pydantic library would.
• The first argument of every call in Annotated metadata now gets inferred a second time, which is unnecessary duplicate work.
• AnnotatedMetadata::callee_name stores just the plain name, so any callee named BeforeValidator gets this pydantic-specific behavior.

IMO a more promising way forward would be to modify just __init__ parameter annotations. Note that direct Annotated[T, BeforeValidator(f)] fields already get an Any __init__ parameter (the reason this doesn't work properly in our tests is the lack of a functional_validators.py stub), so what's still missing is validators behind type aliases.