Add support for more stage-out data types; make xarray optional?

[pont-us]

Currently xcengine only automatically stages out xarray.Dataset instances. We should add other commonly used types. xarray.DataArray is an obvious candidate, but there are also (for example) iris.cube.Cube, pandas.DataFrame, geopandas.GeoDataFrame. This would require a more dynamic data type detection system than the current isinstance(thing, xr.Dataset) to avoid making every potential data library a hard dependency. Such a change could also involve making xarray itself optional -- no sense in having it as a dependency for a notebook that only uses e.g. pandas.

Somewhat related to #61 .

[Hneuschmidt]

Another good candidate would be rasterio's [DatasetWriter](https://rasterio.readthedocs.io/en/stable/api/rasterio.io.html#rasterio.io.DatasetWriter). In the context of Heatwise, we have an example notebook for xcengine that we have to transform to using xarray datasets to work with xcengine.

The link refers to the notebook pre-xarray:

https://github.com/ESA-Heatwise/lstm-wp3-products/blob/51bbc1d5f2755c700f1c5b064addb93096295427/products/hotspots/hotspot_detection.ipynb

Related PR: ESA-Heatwise/lstm-wp3-products#1