This does not make the code simpler but a little more general,
anticipating for example time steps different than one day.

In main program unit `eddy_graph` and procedures `dispatch_snapshot`
and `read_snapshot`, variable k now refers to the date index, 0-based,
instead of the date in days since 1950-1-1.

In main program unit `eddy_graph`, add variable d for date. `k_begin`,
`k_end` and `k_end_main_loop` also now refer to date indices. We have
to pass an additional argument `d_init` to procedure
`dispatch_snapshot`.

Note that no modification is necessary in procedures `get_snapshot`,
`recv_snapshot` and `send_snapshot` although the meaning of k in
`get_snapshot` and tag in `recv_snapshot` and `send_snapshot` also
changes.

We now impose that all eddies in an SHPC have the same
orientation. Remove field `extr_cycl` from the derived type shpc and
add field cyclone to shpc. When reading an eddy, give it the
orientation of its SHPC. Add argument cyclone to procedure
`shpc_create`. Create a file `orientation.txt` in each SHPC. When
writing an eddy, check that its orientation matches that of the SHPC.

Rename triplet to collection, `shp_tr` to shpc, shpt to shpc in
comments, identifiers and file names.

Because variables in `grid_nml` should be all defined in the input file.

We are reverting commits 35fad1b7 and df2b397d because we want the
program `eddy_graph.sh` to read an SHPC containing a single
orientation.

Write `SHPC*` instead of `SHP_triplet_*` directories.

Also split the tests, Fortran source files, Python files and JSon
files. Keep at the top level the files used in both `Inst_eddies` and
Overlap.

Motivations for the split:

- The top directory contained many files.

- It may be useful to compile only one of the two sub-projects. For
example, only `Inst_eddies` on Ciclad, where MPI 3 is not available.

Group `new_unit` and open in the same procedure, that we rename from
`set_unit_edge` to `open_edge_file`. This removes the need for a
special version of `new_unit` and it is closer to the use of the
`new_unit` keyword in Fortran 2008.

This makes the program `eddy_graph` heavier but it makes sense because
there can be no connection between the two graphs, and we need to make
the graphs as small as possible for large spatio-temporal domains.

`unit_edgelist` was just transmitted by procedure overlap from main
program unit to `write_overlap`. So now overlap does not see
`unit_edgelist`. Further motivation: we are going to split the edge
list file into two files.

Add the ability to read a limited number of dates in program
`eddy_graph`. The default is to read all the dates.

Add grid description in a file `grid_nml.txt` in SHP triplets that
should be read by the program `eddy_graph`, instead of reading the
grid description from standard input.

Add test `Eddy_graph_r2md2`.

Replace script `mkdir_nproc.sh` by script
`eddy_graph.sh`. `eddy_graph.sh` is a wrapper script for program
`eddy_graph`. It includes the creation of output directories that was
done by `mkdir_nproc.sh`. `mkdir_nproc.sh` was only used for program
`eddy_graph`. There was the need to automate the concatenation of
output files for a run with several MPI processes.

Add a test for `eddy_graph`.

Improve the handling of the NetCDF Fortran library in cmake files.
Add the configuration of `eddy_graph_in.sh`.

In main program `eddy_graph`, do not write title lines in files
`unit_number_eddies` and `unit_edgelist` because these files are
concatenated for all the MPI processes by `eddy_graph.sh`. So the
title lines are added by `eddy_graph.sh`.

Add test of `eddy_graph` with 2 processes.

Bug fix in main program `eddy_graph`: broadcast `dist_lim` and
periodic.

Bug fixes in main program `eddy_graph`. Add default value for
max_delta. Trim names of files and directories.

Following upgrade of `test_compare.py`, remove `mkdir_run.sh` and
write the mkdir command in the JSON tests files.

Add main program `eddy_graph`.

Change order of dummy arguments of `dispatch_snapshot` and overlap.