Update zonalinfo.py to leave NaN-represented NODATA out of computed averages

[tmcphillips]

When averaging over a region the pixels given NODATA values are ignored during computation when the NODATA value has a numerical value. We need to make sure that this is the case also when NODATA is represented by NaN.

[tmcphillips]

Here is the current situation.

It appears that when NaN is used as the NODATA value for a floating point data set, any averaging over pixels done by zonalinfo.py (called by the timeseries service) returns NaN for the average of any set of pixels that contains one or more NaNs. (This is unlike the behavior when NODATA is represented by some other value. The average returned by zonalinfo.py generally leaves out any NODATA pixels and returns the average of just the valid pixels unless there are no valid pixels at all.)

Superficially, the time series displayed in the web app for LBDA data set, which uses NaN to represent NODATA, appears to be correct for areas. A time series for a region that includes NaNs for a number of years at the beginning simply has no data in the plot (and the CSV download) until the first year in which the averaged pixels all have values. This is the same behavior as for time series taken at a single point (and for a single point this behavior is correct).

The behavior is not entirely correct for areas, however, e.g. when the area covers a region where some but not all pixels have values for the first part of the time series, while all pixels do have values for the remainder of the time series. In this case, the timeseries displayed in the web app has no values at all until the point in the series where all pixels have values. For years in which some of the pixels have NODATA, nothing is displayed even though an average of the valued pixels could in fact be computed.

This problem likely can be fixed by updating zonalinfo.py to leave NaN values out of the averages so that valid averages can be computed and output for any year for which at least one pixel in the selected area has a valid value.

I'm renaming this issue accordingly.