Watermask : implement an alternative to simple NDWI threshold

[ytanguy]

We should add a new option, to filter water based on a NDWI threshold, infered from Pekel occurrences :

• select a Pekel threshold (ex : 80% means 80% of water occurrence in the past 30 years)
• select a percentile (ex : 5%) and compute minimal NDWI (5 percentile) of 80% Pekel occurrence in the target image

We propose to add a new parameter "mode" with different choices (nominal as default) :

• nominal : learn samples selected with Pekel and HAND layers, predict and post-process watermask
• no hand : learn samples selected with Pekel (won't read HAND file), predict and post-process
• relative threshold : read / analyse Pekel and compute a relative threshold on NDWI values, based on % of water occurrence and a percentile (see above), but post-process detection with Pekel
• absolute threshold

This issue is separated in different tasks :

• add new mode parameter and adapt existing modes
• compute a relative threshold to implement the relative threshold mode
• clean and remove obsolete options (ex : Hand strict, area closing, etc.)
• add new tests corresponding to the different modes
• run the test bed with various water bodies and corner cases