Fix deprecation warning from numpy.genfromtxt

[dopplershift]

So this code:

from datetime import datetime

from siphon.catalog import TDSCatalog

metar_cat_url = ('http://thredds-test.unidata.ucar.edu/thredds/catalog/'
                 'irma/metar/catalog.xml?dataset=irma/metar/Metar_Station_Data_-_IRMA_fc.cdmr')
catalog = TDSCatalog(metar_cat_url)
metar_dataset = catalog.datasets['Feature Collection']
ncss = metar_dataset.subset()
query = ncss.query()
query.lonlat_box(north=34, south=24, east=-80, west=-90)
query.time(datetime(2017, 9, 10, 12))
query.variables('temperature', 'dewpoint', 'altimeter_setting',
                'wind_speed', 'wind_direction', 'sky_coverage')
query.accept('csv')
data = ncss.get_data(query)

produces the warning:

python3.6/site-packages/siphon/ncss.py:433: VisibleDeprecationWarning: Reading unicode strings without specifying the encoding argument is deprecated. Set the encoding, use None for the system default. converters={'date': lambda s: parse_iso_date(s.decode('utf-8'))})

For some reason, our existing tests don't trigger this warning. The warning occurs because numpy 1.14 started being more pedantic about the encoding stuff, and adding an encoding keyword. We can make things work fine by passing this kwarg with a value of 'utf-8', but of course that means we only work on numpy 1.14. Switching to StringIO instead of BytesIO does not solve the issue. Options I can think of:

1. Bump numpy requirement to 1.14
2. Add our own wrapper/helper for genfromtxt that is a straight passthru on numpy > 1.14, and drops the encoding kwarg otherwise
3. Add a warning filter to swallow the deprecation warning, bump numpy to 1.14 much later.