UUID will be easy as a logical type on top of fixed size binary physical types.
I believe for float16, we were considering adding it but only held off because of the lack of kernel support for it from arrow2. Happy to revisit though.
As for the large_* variants, we default to using 64bit offsets which are the arrow large_* variants. This is an implementation detail of of the types though, eg we will at some point move over to something such as Umbra string representation vs the current arrow large_string representation for our string arrays.