Propagate change in shapelib_03: shp_append_simple_object_03 to
shp_append_object_03.

namelist input. Also, choose name of primary NetCDF variable in
namelist and read only this variable, not its coordinates. This makes
it easier to use an arbitrary NetCDF file.

In program test_set_outermost_contour, it is hard to remember the
order of positional arguments so we rather use fixed input file names
and no arguments.

In JSON input file for tests, it is simpler to use the "input" key
than creating a one-line input file, in some cases.

out-of-range index so add min when calculating urc.

Bug fix in procedure good_contour. polygon_contains_point should not
be called if polyline is not closed.

Procedure polygon_contains_point became unpure so remove pure for
procedure inside_4 too.

In program test_good_contour, read corner and step from namelist
rather than from coordinates in NetCDF file. Makes it possible to use
the same NetCDF file with different coordinates. Also read
outside_points from a file rather than standard input so there is no
need to specify in advance the number of outside points.

Take into account the possibility that no outermost contour was found
in program test_set_outermost_contour. Also, use fixed names for input
files because it is not convenient to require a particular order of
these two files on the command line.

Make program test_good_contour more general.

outermost_possible_level. This is especially important if the extremum
is one point away from the last point.

Bug fix. In procedure set_outermost_contour, take level_good a little
bit closer to ssh_extremum than innermost_level because of border
points. Contour_531 will not close a contour passing at a point on the
border. So if innermost_level corresponds to a point on the border,
good_contour at innermost_level finds nothing.

Make the program test_good_contour more general: read a NetCDF file
instead of calling define_example_reg.

test_good_contour has more general input.

test_inside tests only inside, no loop on radius.

procedure get_eddy), with corresponding test.

Take into account the fact that derived type polygon became polyline
in library Contour_531.

Use new procedure null_polyline from library Contour_531.

Correct a sneaky bug in mean_speed: the last point in the polygon
array is a duplicate of the first point, do not take the last point
into account when averaging (and interpolating). Discovered this bug
because the mean speed was not exactly the same for equivalent
polygons with circular permutation.

Change philosophy for the tests: read whole NetCDF variables, prepare
different NetCDF files for different geographical regions if needed;
read all data from file or namelist; read names of files on command
line.

Use generic test_program.sh (which is out of this project) instead of
the specific tests_detection_eddies.sh.