We are circumventing a bug of the Intel Fortran compiler. This was
tested with both ifx `2023.1.0` and ifort `2021.9.0`. The bug occurs
on return from procedure `set_max_speed`: `free(): double free
detected in tcache 2` and crashes execution. The bug is revealed in
commit cc13fd5b. The GNU compiler does not have this bug.

To circumvent the Intel compiler bug, instead of an ssh component of
`cont_list_proj`, we have to use a separate array:
`cont_list_ssh`. This is maybe for the best because the component
`area` of `cont_list_proj` was not defined.

`n_points` can be deduced from previous print, and `n_extr` has been
printed too.

If `l != 0` then `shape_rec.record.r_eq_area >= 0` so `n_points` was
always equal to `n_points_valid`: no need for a separate variable.

Because the loop on shapes may take long.

`MAX_RADIUS_deg` is not general enough for a non longitude latitude
grid. The variable `max_radius` moves from module `input_ssh_m` to
module `config_m`. The variable `MAX_RADIUS_deg` disappears. The
constraint `2 * max_radius_deg(1) < 180` becomes `4 * max_radius(1) <
nlon` but is only applied to the periodic case because we do not have
the general equivalent of 180° in number of points.

In procedure `get_var`.

Promote nlon and nlat to variables of module `input_ssh_m`, avoiding
their repeated computation.

Preparing to allow access to variables of module `input_ssh_m`.

Not worth it for a single point.

Rename variable `outside_points_proj` to `outside_points`.

`Region_1_20454_cyclone_3/config_nml.txt` is a symlink to
`Extraction_eddies_region_1_noise/config_nml.txt`.

In main program units `inst_eddies`, `examine_eddy` and
`test_get_1_outerm`, redefine `outside_points` as coordinates in
projection space, avoiding a round trip to longitude and latitude. In
`examine_eddy`, we therefore store `coord_proj` instead of coord.

From `nearby_extr`.

Instead of `outside_points`.

Instead of `outside_points`.

Avoiding conversion back and forth to real.

Instead of corner.

Instead of corner.

Redefine `corner_window` as coordinates in projection space, avoiding
a conversion back and forth to longitude and latitude.

Instead of corner.

Using the fact that `(corner_whole - corner) / step` is mathematically
an integer.

Instead of from `out_cont`.

After reading `cont_list_proj`.