Handle large area processing

[ValentinaHutter]

odc_load_helper now only changes nodata values to np.nan

[sophieherrmann]

There are some nice improvements!

Did you already check all other functions if attributes are properly passed?

[sophieherrmann]

I just found this http://xarray.pydata.org/en/stable/generated/xarray.save_mfdataset.html

could improve speed of writing netcdf files

[ValentinaHutter]

I just found this http://xarray.pydata.org/en/stable/generated/xarray.save_mfdataset.html

could improve speed of writing netcdf files

Thanks, I just inserted that :)

[clausmichele]

Why are you splitting the request into two separate ones? I mean, this should be more flexible and split it into parts which have a size you know it's fine, otherwise this could work for size x = 100 for example, where you split into x1 = 50 and x2 = 50, but what if x = 1000? (numbers are just examples)

[ValentinaHutter]

Thanks for reviewing it, that was just a test to try out splitting it in two parts, but I will remove the change, as it is not improving the process.

[ValentinaHutter]

To handle large areas and apply processes to large areas, I had a look at all the processes. The processes that still need an update to work for large areas are sort, order. The issue is discribed here: #52

[sophieherrmann]

As this PR already provides a number of new features / bug fixes which are urgently needed, I'll merge it now.
The main new feature is, nearly all jobs are now running completely on dask - no direct access to the array values. This allows to also run large area jobs. Exceptions are described by @ValentinaHutter #49 (comment) and will be solved in a separate PR.

A connected issue is that the fit_curve process is computationally quite expensive (especially for large areas). This issue is documented here #53 and will also be addressed in a separate PR.