new low memory version

[esheldon]

This implements a low memory footprint version of the code

This version uses the following additional storage, beyond that in the original ra,dec arrays

• storage for the matches if not writing to a file
• possibly byteswapped versions of ra,dec for both catalogs if they were big endian (e.g. if you read from a fits file). You can save mem by doing that yourself (e.g. esutil.numpy_util.to_native(ra, inplace=True)
• temporary storage of x,y,z for the second catalog

[esheldon]

@erykoff check it out