Several problems with watermask when Pekel is empty

[ytanguy]

When Pekel mask is empty (no significant water area known in Global Surface Water), SLURP 0.1 shows several problems :

• it turns into "NDWI threshold" mode but the threshold is hard coded (1000) instead of args.ndwi_threshold
• the "no_pekel_filter" option should be switched to True, otherwise all filtered areas will be discarded because Pekel mask is empty !
• finally, the morphological operations are made in a wrong order and a lot of false positive are detected.

The order of post-processing operations should be :

• remove_small_objects
• remove_small_holes
• closing
• opening