Fix resolution of forward refs in dataclass base classes that are not present in the subclass module namespace

[matsjoyce-refeyn]

Change Summary

(Text mostly copied from #8730) When the base class of a dataclass has an annotation type that is not imported by the module the subclass is in, and from __future__ import annotations is in use (or in my case, Cython is in use), an error occurs.

The root cause is that when pydantic builds up the type namespace, it does not consider base classes (https://github.com/pydantic/pydantic/blob/main/pydantic/_internal/_generate_schema.py#L1474). This is fine for models, because each base class builds a schema immediately, but does not work for plain stdlib dataclasses. This PR adds code to iterate though all of the base classes of the dataclass that are also dataclasses, and push their namespaces as well, which should cover all of the fields which Pydantic will attempt to resolve.

Related issue number

Fixes #8730

Checklist

• The pull request title is a good summary of the changes - it will be used in the changelog
• Unit tests for the changes exist
• Tests pass on CI
• Documentation reflects the changes where applicable
• My PR is ready to review, please add a comment including the phrase "please review" to assign reviewers

Selected Reviewer: @alexmojaki

[codspeed-hq]

CodSpeed Performance Report

Merging #8751 will not alter performance

_{Comparing matsjoyce-refeyn:bugfix/pydantic-8730-fix-base-dataclass-forward-refs-unknown-error (daa2ba8) with main (825a692)}

Summary

✅ 10 untouched benchmarks

[matsjoyce-refeyn]

The only job failing on CI is the tags check, so please review

[alexmojaki]

Thank you, this looks promising!

[matsjoyce-refeyn]

All comments have been addressed, please review (looks like hooky doesn't react to rereview requests).

[alexmojaki]

Great, thanks!

[matsjoyce-refeyn]

@alexmojaki Just to check, is there anything else I need to do to get this merged?

[sydney-runkle]

I've tested out using swapped orders for traversing the MROs, and I think the current approach makes sense. There were test failures for all other options re MRO traversal @dmontagu