Support UUID as arrow type

[eddyxu]

• Support specify UUID as arrow schema
• Stored as fixed size data type (128 bits)
• Have Python and Javascript definition in both LanceDB SDKs

[rok]

Is this meant as extension type (e.g. https://github.com/apache/arrow/pull/37298) or as part of lance format?

[eddyxu]

It'd be great that we can natively support UUID type from pyarrow. ALso we wanted to support the interoperability in javascript as well.

[rok]

I have an upstream PR in arrow that may get merged for 15.0.0 apache/arrow#37298.
Should we wait for one quarter to get it or implement a temporary equivalent in pylance (e.g. #1471)?

I'll start working on the js one.

[rok]

@eddyxu I was planning to continue work on this after the upstream extension type was merged (apache/arrow#37298). If you'd want UUID type in Python before that #1471 can be merged. If not #1471 can be closed out.

[eddyxu]

Once we have upstream PR merged, do we need anything change from lance to support it?
I can take it over from there.

[rok]

It'll be available in pyarrow as e.g. FixedShapeTensorArray is, see test here. I'm not sure if you'll need changes on rust side of things.