PySpark codegen silently falls back to StringType for unknown scalars

Summary

Types whose validation is supplied by pydantic-core are only covered when the
codegen explicitly implements them. Anything else extracts as a Primitive with
an unregistered base_type, silently falls back to an unchecked StringType(),
and gets zero checks. This also counts for pydantic-extra-types , Overture schema
and 3rd party scalars.

Affected fields today:

• sources[].update_time: datetime | None
• annex inception_date: date | None

These are admittedly multiple issues:

• No transparency: the fallback is invisible (logging)
• In-use scalars are missing (datetime/date)
• Most pydantic ecosystem's (core & extra) scalars are missing
• No registration mechanism for third-party scalars