add varlist argument to dfmt.get_Dataset_atdepths() or move layer vars to grid-like acessor

[veenstrajelmer]

With large datasets the dfmt.get_Dataset_atdepths() function is very computationally expensive. Providing a subset of the dataset is possible, but the user should be aware of which layer variables should be passed on. This is inconvenient. In the below example only mesh2d_ucx and mesh2d_ucy are needed by the user, but many variables are required for the depth slicing:

import dfm_tools as dfmt
file_nc = r"p:\11209731-002-nutrient-reduction-ta\runs_OSPAR\B05_waq_withDOM_2016\DFM_OUTPUT_DCSM-FM_0_5nm_waq\DCSM-FM_0_5nm_waq_0000_map.nc"
uds = dfmt.open_partitioned_dataset(file_nc)
uds_sel = uds[[ 'mesh2d_ucx', 'mesh2d_ucy', 
               'mesh2d_layer_sigma', 'mesh2d_layer_z', "mesh2d_layer_sigma_z", "mesh2d_interface_sigma_z", 
               "mesh2d_s1", "mesh2d_bldepth", "mesh2d_sigmazdepth", "mesh2d_interface_z", "mesh2d_interface_sigma",
               ]]
uds_sel_depths = dfmt.get_Dataset_atdepths(uds_sel, depths=[-1])

Desired solution
Add a variables_toslice argument to dfmt.get_Dataset_atdepths(), in this case variables_toslice= ["mesh2d_ucx","mesh2d_ucy"]. If provided, this would overwrite the derived variables_toslice in the function. It is derived based on time/faces/layers dimensions, so when passing a 1D variable, it will also be sliced (eg bedlevel will also get several layers).

Alternative
It might be better to extract the layer data out of the dataset (like the .grid data is also kept separately). This would also likely solve things for d3d4, schism, rst and other formats.