The Juwels Booster configuration is a reduced set of variables only
suitable to perform remapping. The extraction of CMIP6 data has to be
done at DKRZ.

The CMIP6 data is not available in Jülich. The configuration merely
covers settings for interpolation.

When choosing different CMIP scenarios a set of variables have to be
consistent. These are now set based on the chosen emission scenario.

As a side effect, all AMIP years now have full neighboring years.

Add the `-L` option to CDO as the netCDF/HDF5 library is not thread-safe.

Apparently some ICON version require at least 14 months input.
Some other versions do not. Who am I to judge?

Do not adjust the mean SST to the 2020 average value. Instead, use the
SST and SIC values from individual years and only shift the time axis.

This leads to a better representation of the internal variability of the
climate system, and more consistent SST and SIC data sets.

The old implementation used the lowest atmospheric geopotential level.
However, this is not the surface. Instead, use the orography and
convert it to gepotential.

Include all grid information in weights file.

Clean-up configuration file.

Also increase initial atmospheric temperature data. Use proper `aexpr`
operator to include all input variables and only append increased
temperature data.

Fix units for SST (in K) and SIC (in range [0 - 1]). Also change long
names to be consistent with ICON direct BC input.

Apply the running mean auotmatically after extraingt the data from CMIP.
Adjust the averaging window based on the temporal resolution.

Do not hard-code the year range but determine it based on the initial
date.

Make the institution that performed the simulation variable.

ICON expects the cell dimension for IFS input files to have a certain
name. Rename it for specific input files.

This approach is more stable between different hosts

* Fill missing values **before** remapping
* Enable extrapolation for grid points "outside" coarse source grid