Useless in this script.

Simply printing g displays the address of the graph, which may change
between runs. This was not convenient for the comparison of runs.

To ease comparison of runs.

Do not assume that each SHPC corresponds to a given year. Do not
impose any directory structure on the input SHPC. We have to bisect
through the list of `d_init` to find the SHPC containing a given date,
as in program `eddy_graph`.

Since we no longer load all the contents of shapefiles for a given
year in advance, we no longer need to process the segments year by
year. Motivation: this will allow us to get free of the imposed
arborescence of SHPC by year.

Create properties in `cost_function.py` rather than in
`segments.py`. We do not use these properties in `segments.py`.

Use `comp_ishape` and `open_shpc` from `util_eddies`.

Do not store in memory all the records from shapefile
`max_speed_contour`.

But we should find some way not to impose a directory structure to the
user.

We do not want hard-coded paths in the script.

We do not want to change the script to change orientation.

Assume that `node_id_param.json` is in the current directory. We do
not want paths written inside the script.

With no `d_init` argument. In the returned value, we use key
`date_index` instead of `days_1950`.