Support serialization of query results to in-memory arrow data types about kuzu HOT 1 CLOSED

To add more detail (I have a partial implementation which isn't quite working yet), it looks like for the C++ API, if we want to move away from depending on arrow we can just expose the ArrowArray and ArrowSchema types from the C Data Interface (the backend for the python API already produces those types prior to converting them into a pyarrow table, that code just isn't exposed in the releases at the moment). They should be together functionally equivalent to RecordBatch, and I think they can be converted more or less directly to a RecordBatch (not sure which exact calls are required).

In the Rust API on the other hand, ArrayData can be constructed from the ArrowArray and ArrowSchema over the FFI, and then can be converted directly to StructArray and then RecordBatch. We could probably even do something similar to C++ on the rust side and keep the dependencies to a minimum by just producing arrow_data::ffi::FFI_ArrowArray and arrow_schema::ffi::FFI_ArrowSchema, which would remove the need to depend on the main arrow crate, but I'm not sure if the arrow dependency is as much of an issue there.

from kuzu.