Improve reading model output

[dbrakenhoff]

Use dask.delayed to load model output. This avoids reading all data into memory if delayed=True. Optionally chunk data array for doing memory efficient calculations on large data arrays.

Adds two kwargs to output methods:

• delayed, if True, do not load data into memory, default is False
• chunked, if True, chunk data array using da.chunk("auto"), default is False

Default behavior is same as before, for memory intensive output, use delayed=True, and optionally chunked=True, e.g.:

heads_orig = nlmod.gwf.output.get_heads_da(ds)  # read all data into memory
heads_delayed = nlmod.gwf.output.get_heads_da(ds, delayed=True)  # memory efficient
heads_chunked = nlmod.gwf.output.get_heads_da(ds, delayed=True, chunked=True)  # chunked

[OnnoEbbens]

Good job!

I was thinking if we should make this functions cache-able using the cache_netcdf decorator but I think it is too hard (and to little gains) because the cache_netcdf requires a dataset as input and for this function it is optional. However if these functions do become annoyingly slow we could evaluate again because it is certainly possible to use the cache here.

[dbrakenhoff]

Alright, the idea is still the same but I refactored the code significantly to increase readability and simplify things.

The idea now is that the binary output file (HeadFile and CellBudgetFile) and the modelgrid should contain enough information to construct a DataArray. If you pass in only a filename to e.g. get_heads_da(fname=fname) you will receive a warning that the grid information is missing. This information can be provided by passing grbfile=<path to binary grid file> keyword argument. You can still load data, but the grid will be some default grid definition (and will not contain the correct spatial coordinates).

• The file mfoutput/mfoutput.py contains general logic for converting data from flopy binary file objects (HeadFile and CellBudgetFile) to data arrays. I defined a bunch of helper functions to reduce duplication and keep functions short.
• The file mfoutput/binaryfile.py contains code to read data from binary output files but supporting multithreading. Code is copied from flopy, but modified to contain only the necessary code. We do not support the same level of data accessing options as flopy.
• gwt/output.py has been modified to use these new general functions for concentration data
• gwf/output.py has been modified to use these new general functions for head and budget data

[OnnoEbbens]

see my comments