Refine NOBS calculation in LIS v1.11 snow synthesis

Right now the NOBS calculation includes, for a given pixel, all the clear observations provided between start_date - margin and end_date + margin, meaning that multiple observations from within the margins can be taken into account. Though only the latest one (for the start of the time period, resp the first one for the end of it) is really used later on to strengthen the interpolation over the time window boundaries.

Wouldn't it be more accurate to count, for a given pixel, all the clear observations between start_date and end_date, then add 1 if at least one clear observation is found for that pixel in the [start_date - margin, start_date] time period, then add 1 again if at least one clear observation is found for that pixel in the [end_date, end_date + margin] time period. That way, only the observations really used during the interpolation would be counted in the NOBS.

There would be little change for a short margin value, but given the recommended margin is one month, and the default margin is two weeks, it could possibly increase the NOBS by around 10%.