The fact that performing the sums in different order still induces differences but they are small.

The addition of a landsendtohalo to fetch_basin does not work. Leads to a lock-up in truncate.

But there still seems to be an issue with the calculation of the fetch.

Prepares and tests functions to echange halo and make some operations on core points only (zero and sum).

Corrected some stuff and ensured that the diagnostics from the F90 part are now written into files. This allows for finer diagnostics of the computation there.

They are not yet adapted to the parallel processing.

Also some small corrections on the parallel processing.