feat: Pydantic preprocessing and interfacing to NanoEventsFactory

[NJManganelli]

This continues the series of PRs to make pydantic input models thread through coffea, and specifically enables preprocessing parquet files via this. The old preprocess is renamed preprocess_legacy, and preprocess now attempts to coerce users to pydantic classes (with an escape hatch, preprocess_legacy_root=True). The pydantic preprocess is made a nonpublic function, with two user-facing variants for root and parquet which call it appropriately.

This partially pulls in updates from https://github.com/NJManganelli/coffea/tree/datafactory_parquet and some associated prototyping in another branch which did not get incorporated into #1403

This walks back changes to preprocess, to make the code friendlier and make it easier should we deprecate the dict-based preprocess.

More to be edited into the description as it evolves.