time splitting and isf: bugs

[mathiot]

Context

In the trunk, 3 bugs were found recently with time splitting and ice shelf:

• bug1: the mask used to compute the average ht_n was wrong (0 under ice shelf). It affects run done with isf+time-splitting+een. Tests with ISOMIP show a negligeable impact (difference of ~1e-7 in ssh_max after 30 years). Corrected in dev_merge_2017 #9546 (gm)

• bug2 :the mask used to compute the new depth at velocity point was wrong (0 under ice shelf). It affects run done with isf+time-splitting+flux-form. Test with ISOMIP show a 10% decrease of moc and stream function below the ice shelf when the bug is corrected (see attachment). Corrected in dev_merge_2017 #9554

• bug3 :the mask used to mask ubaro is wrong (0 below an ice shelf). It only affect the output. Corrected in dev_merge_2017 #9554

In version 3.6_stable, the use of time-splitting with ice shelf is not possible (protected by a stop).