Datacube.load should better support CRS object and atemporal products

[benjimin]

Support for atemporal products (such as a static DEM) should be improved, particularly in Datacube.load as in this example where the like parameter behaves unexpectedly. Also, the same function should fully support the datacube.utils.geometry.CRS object, rather than requiring e.g. WKT conversion for some fields and not others.

Obvious use case: extract any limited xarray from a datacube, then seek to combine with DEM.

Setup:

lat, lon, e = -35.09407, 150.80426, 0.01
import datacube
dc = datacube.Datacube()
x = dc.load(product='ls8_nbart_albers', time=('2019-06', '2019-07'), 
            lat=(lat-e,lat+e), lon=(lon-e,lon+e))

Problem 1: This should return data:

>>> dc.load(product='dsm1sv10', like=x)
<xarray.Dataset>
Dimensions:  ()
Data variables:
    *empty*

Problem 2: This should completely work:

>>> cr = lambda coord: (coord.data.min(), coord.data.max())
>>> dc.load(product='dsm1sv10', x=cr(x.x), y=cr(x.y), 
...         crs=x.crs, output_crs=x.crs, resolution=(25,-25))
TypeError: unhashable type: 'CRS'

The combination of the problems: the following does work and is confusingly cumbersome:

>>> dc.load(product='dsm1sv10', x=cr(x.x), y=cr(x.y), 
...         crs=x.crs.wkt, output_crs=x.crs, resolution=(25,-25))
<xarray.Dataset>
Dimensions:    (time: 1, x: 86, y: 100)
...
Attributes:
    crs:      EPSG:3577

[Kirill888]

@benjimin open time ranges is what needed in this case. The reason why like doesn't work in this case is that

1. It injects time=() into a query, because input has time axis
2. We do not have open ended ranges so we can't mark "atemporal" datasets with time=(0, +inf)

Work around for your case is to remove time dimension when using like= parameter:

>>> dc.load(product='dsm1sv10', like=x.isel(time=0))

[Kirill888]

Also see #808, when merged you'll be able to do:

>>> dc.load(product='dsm1sv10', like=x.geobox)

[benjimin]

Yep, query by geobox is helpful. (Had totally forgotten I'd offered a PR for that years ago.)

Still, crs argument of Datacube.load should tolerate a datacube CRS object, don't you think?

[Kirill888]

@benjimin I missed that part, can you make a new issue for that, certainly output_crs= can/expects CRS object, I assumed same applied to crs=

[Kirill888]

So the CRS part is fixed by #812. Hopefully #808 helps with atemporal products. And since we already have #183, I'm going to close this issue.