Improve cloudmask legend

[jordancaraballo]

Would be great to improve the way we present the cloud mask. This way we could have a way of stating pixels that we just did not properly classify based on confidence.

Class	Description
0	Not cloudy
1	Cloudy
2	Not cloudy but adjacent to a cloudy pixel
200	Could not be classified reliably

[pahbs]

during this improvement we might consider testing on L8 and S2 data.

Most importantly, we should be think about a legend that is consistent with other cloudmasking approaches: https://www.sciencedirect.com/science/article/pii/S0034425722001043

[jordancaraballo]

With the new contours we can have 2 - not cloudy but adjacent. Need to implement in the next sprint.