ENH: Support for xarray

[jklenzing]

Linked from pysat/pysat#47.

Current seasonal routines assume pandas data. This needs to be generalized to accept xarray data.

[jklenzing]

Once generalized, xarray can be used to replace Panel, addressing #4.

[jklenzing]

Proof of concept:

import pysat
import numpy as np
lats = np.linspace(-90, 90, 20)
test_inst = pysat.Instrument('pysat', 'testing_xarray')
test_inst.load(2009, 1)
test_inst.data['latitude'].groupby_bins(test_inst.data['latitude'], lats).mean('time')

produces

<xarray.DataArray 'latitude' (latitude_bins: 19)>
array([-8.68204547e+01, -7.60356830e+01, -6.64236810e+01, -5.68964574e+01,
       -4.74291039e+01, -3.79554216e+01, -2.84429217e+01, -1.89403821e+01,
       -9.45146641e+00, -1.50410421e-15,  9.45146641e+00,  1.89403821e+01,
        2.84429217e+01,  3.79554216e+01,  4.74092899e+01,  5.68964574e+01,
        6.64236810e+01,  7.60356830e+01,  8.68242926e+01])
Coordinates:
  * latitude_bins  (latitude_bins) object (-90.0, -80.526] ... (80.526, 90.0]

Needs to be generalized for N dimensions. Looking through old stack overflows, this is not yet supported in xarray, but we may need to dig through the more recent documentation.

https://stackoverflow.com/questions/40465026/groupby-bins-on-two-variables

[rstoneback]

I have a potential method that scales to ND over in #22. I'm planning on including these very topics in the upcoming proposal to NASA.