When there are more than 3 input files. We recover the convenience of
commit b34b85a8, with the underlying generality of commit d365e754.

So we get the function without executing the rest.

So we can call it from another script with a generated list of files.

Because we are going to extract the rest of the script below into a
function.

Check we find a max-speed contour when `e%radius4 >= 2` and
`e%speed_cont%ssh /= missing_ssh`. See test `Set_max_speed_degenerate`
for a case when we do not find one.

And it is then conveninent to choose it from the command line.

So read `grid_ml.txt` at the beginning of the script and not in
function `plot_grid_bb`, and pass `grid_nml` instead of `shpc_dir` to
function `plot_grid_bb`.

The script `plot_eddy_contours.py` was becoming too complicated.

If `n_dates` has been defined in the namelist then it is used by
`read_column` and, after the call to `read_column`, it is already the size
of `ishape_last`. It is clearer not to redefine it.

Use separate directories for modules of each target to circumvent
[Issue 22898](https://gitlab.kitware.com/cmake/cmake/-/issues/22898)
of CMake.

Write input in json file instead of preparing a separate input file.

Provide list of input files as arguments of `inst_eddies.py`,
following commit 8e0c03e1.

For programs `inst_eddies` and `test_get_1_outerm`, following commits
9a59d13 and 9dc139d.

Following updates of the libraries.

Following commit b0e3ce55.

The programs in this project do not depend directly on them. If they
are secondary dependencies, it is the responsibilty of the primary
dependencies to find them.

As for the program `inst_eddies`.

We want to generalize our possible sources of NetCDF files. We no
longer make assumptions of their filenames. And the time step may not
be the day so we cannot infer a timestep number from the filename. So
we just read a list of files as arguments and we read the first date
from standard input. We pass the date to standard input of the Fortran
executable.

Do not print default namelist, since we no longer enter the namelist
on standard input.

Read the namelist from a file instead of standard input because now we
read `days_1950` from standard input in the main program unit. When
`inst_eddies` is called by `inst_eddies.py`, `days_1950` has to be
passed at each run of `inst_eddies` by `inst_eddies.py` while the
namelist comes from out of `inst_eddies.py`. So we have to separate
the two input fluxes.

Start generalizing the processing of time: do not read time from
NetCDF file. We no longer assume that the time coordinate in the
NetCDF file is in days since 1950-1-1 and that the time step between
NetCDF files is one day. So the program `inst_eddies` can no longer
know what the date is, it just asks it to the user.