A big-endian data variable registers, then every query fails: Byte-swapped arrays not supported #288

A Dataset whose data variable is big-endian, as a NetCDF file written on a big-endian machine, or some FITS or binary files, can be, registers without error, and then every query on it fails.

import numpy as np, xarray as xr, xarray_sql as xql

ds = xr.Dataset({"w": (("a",), np.zeros(3, ">f8"))}, coords={"a": np.arange(3)})
ctx = xql.Context()
ctx.from_dataset("t", ds)
ctx.sql("SELECT * FROM t").to_pandas()

On main:

DataFusion error: FFI error: Execution error: Error reading batch from stream: ArrowNotImplementedError: Byte-swapped arrays not supported

pyarrow can't make an array from a byte-swapped NumPy array. With #275, such a Dataset is read lazily, as on main, so it fails the same way.

Expected: the query works. One fix is to convert the values to the machine's byte order (values.astype(values.dtype.newbyteorder("="))) when making a block's Arrow column. That costs a copy, which an array in the machine's byte order doesn't need.