The bug comes from commit 06c451e2.

Read date from `date_range.txt` in program `inst_eddies` and update
date in `date_range.txt`at the end of execution of program
`inst_eddies`, remove date from the namelist `main_nml`.

This is a change in the format of an SHPC. I have realized that the
program `inst_eddies` may not find any eddy at a given date and
orientation (as in the test `No_eddy`), so we cannot guarantee that
the first date present in the shapefiles corresponds to the first line
in `ishape_lat.txt`. So we store the initial date explicitly in a text
file. We could deduce the final date from the initial date and the
number of lines in `ishape_lat.txt` but we can avoid counting the
lines and just store the date range.

Preparing to only append to shapefiles in the Fortran program, not
create them.