Else the output of `subprocess.run` comes before the print statements
in the script.

Clarifying the logic: the Fortran program only appends to an already
created SHPC slice.

Use `create_slice.py` in tests of program `inst_eddies`. The namelists
then have to be written to a file, to be read both by
`create_slice.py` and by the program `inst_eddies`.

The program `test_mean_speed` no longer needs the SHPC directory since
commit 8eadc458.

Call `create_slice` from script `inst_eddies_INALT.py`, instead of
just removing and creating the directories. Note that we then need to
know `uniform_lon_lat`.

Instead of just removing and creating the directories. Note that we
then need to know `uniform_lon_lat`. Also note that we need to copy
the file `create_slice.py` to the build directory where it is used by
`inst_eddies.py`.

Preparing to only append to shapefiles in the Fortran program, not
create them.

Write longitude and latitude if `leave_trace` and not `uniform_lon_lat`.

A negative value is even more striking since a step for a uniform grid
would be positive.

Do not try to plot grid bounding box if not `uniform_lon_lat`, for
now.

Only write `used_config_nml.txt` and `grid_nml.txt` if `leave_trace`.

Promote u and v to variables of module `input_ssh_m`, and remove them
as dummy arguments of `input_ssh`, `set_contours`, `mean_speed` and
`set_max_speed`. u and v are defined in procedure `input_ssh` and used
only in procedure `mean_speed`.

Remove dummy argument corner of function `mean_speed`, not used any
longer.

The case of u and v is different from the case of ssh. For ssh, we
want to restrict as much as possible the size of the array we work
with because we search all the contours in the corresponding
window. But for u and v, we only do a bilinear interpolation, without
a call to locate or hunt within the interpolation procedure, so there
is nothing to be gained to work on only a window of u and
v. Furthermore, we are preparing to allow staggering of the u and v
grid with respect to the ssh grid so if we wanted a window of u and v,
there would be the complication of choosing the window taking the
staggering into account.

It was not convenient to read them with input function or on the
command line: too much to read this way, and not clear when saved in
JSON text file. And we already had to create a namelist file
anyway. And this way the namelists can be extended transparently.