Conversion of numpy boolean array to pyarrow boolean array isn't zero-copy

[sjperkins]

In both NumPy and CASA, booleans are represented by a byte (8-bits). However, Arrow represents booleans with a single bit.

This means that, in the case of np.bool_, this conversion is not a zero-copy:

arcae/src/arcae/lib/arrow_tables.pyx

Lines 242 to 243 in 8d1da1b

	# Convert to pyarrow array
	pa_array = pa.array(np_array)

This means that it's not possible pass NumPy bool arrays as getcol result arguments, or as putcol data arguments.

It's easy enough to work around this by passing data.astype(np.uint8) in these instances, as arcae represents CASA booleans as arrow:uint8()

arcae/cpp/arcae/type_traits.h

Lines 58 to 64 in 8d1da1b

	template <>
	struct CasaDataTypeTraits<casacore::DataType::TpBool> {
	using ArrowType = arrow::UInt8Type;
	using CasaType = casacore::Bool;
	static std::shared_ptr<arrow::DataType> ArrowDataType() { return arrow::uint8(); }
	static constexpr bool is_complex = false;
	};

but it'd probably be convenient for the user to do this within arcae itself, during the numpy <--> arrow conversion process.