Note that this is still pretty experimental, and all plugins still pin pydantic v1. I'd like the version loosening to go in the 0.12.1.2 release.

If there's issues with our pydantic usage / configs, pydantic throws errors at class creation time. That means that checking datahub check plugins should largely help us check if stuff will work. Additionally, all the tests in tests/unit/config pass with both v1 and v2.

As of now, GX, feast, and a few other sources hard require pydantic v1. Our main CI tests still install pydantic v1 because it's the lowest common denominator. I did upgrade the Airflow 2.7 test to require pydantic v2. However, it's still not a great way of ensuring that we're compatible with both versions.

The data contracts CLI also continues to use pydantic v1 APIs (with either v1 or v2), since it's not easy to write code that works with both v1 and v2 versions of root models.

Closes #8829, follow up on #9220.

Follow ups

• Add matrix tests for both pydantic v1 and v2
• Remove pydantic v1 as a requirement for most sources

Checklist

• The PR conforms to DataHub's Contributing Guideline (particularly Commit Message Format)
• Links to related issues (if applicable)
• Tests for the changes have been added/updated (if applicable)
• Docs related to the changes have been added/updated (if applicable). If a new feature has been added a Usage Guide has been added for the same.
• For any breaking change/potential downtime/deprecation/big changes an entry has been made in Updating DataHub