Feature request: XArray backend

XArray provides a pandas-like interface to regularly gridded (with missing values) data. This has a very nice analogy to the ds.r[] object, when it is sliced with imaginary step sizes -- i.e., covering grids or arbitrary grids. It would be excellent to provide either a conversion from a covering grid to an xarray or to make this a returned output from some functions.

While it is feasible to do this as-is, for instance:

ds = yt.load("IsolatedGalaxy/galaxy0030/galaxy0030")

d = ds.r[::512j, ::512j, ::512j]

density = d["density"]
coords = [('x', d["x"][:,0,0]), ('y', d["y"][0,:,0]), ('z', d["z"][0,0,:])]
rho = xr.DataArray(density, coords=coords)

This requires that the data be loaded in all-at-once. XArray provides the ability to defer loading, either by providing something as a dask array or by implementing a backend.

Implementing a backend seems to be the most straightforward way, as it would require subclassing AbstractDataStore, perhaps in the mode of the opendap backend. This would be a very cool way to make yt more interoperable with xarray.