provide timing statistics without first and last time steps
Timing statistics on the performances of each routines are performed on the whole code, including the initialisation phase, the first time step which contains some initialisations and control pint and the 2 last time steps which usually deal with the restart files.
The poor/weird performances of the initialisation/restarts phases can be quite misleading. I therefore duplicated all the timing reports, the first one based on the entire simulation, the second one based on the timestep nit000 + 3 * nn_fsbc
to nitend - 2 * nn_fsbc