Plotting logic

[darothen]

Here's an initial port of the custom pcolormesh code. It needs a re-factor, but I include a sample script which uses the test data from NASA directly to plot.

To-do

• [ ] Re-factor grid-specific (face order/rotations) details out of pcolormesh
• [ ] Re-factor polygon computation to separate routine
• [ ] Add sketch of xarray accessor

[JiaweiZhuang]

Thanks! The plot looks great! Several questions before merging:

1. Any reason to prefer shape (lon, lat, 6) for both grid calculation and plotting? (6, lat, lon) seems more natural to me, as the tile dimension should be the slowest-changing index.
2. face_wrap = True breaks the plotting function because I didn't import dependencies for calc_ϕ_on_great_circle. But I don't notice anything weird with face_wrap = False, so I wonder in which case face_wrap = True is needed?
3. Any way to hide gridlines? The edgecolor keyword doesn't seem to work.
4. Any room for speeding up the plotting function? (C96 plot takes 50s on my Mac...) The speed of grid calculation isn't crucial because it only needs to be done once, but users will call plotting all the time. You don't have to do it now. Just wondering.