Change calculation of _REF variables to not use values in long gaps

[gilbertozp]

Calculation of _REF variables (for NEE, GPP, and RECO), for all resolutions (HH, DD, WW, MM, YY), in nee_proc and ure steps, currently use gapfilled values in the calculation of the model efficiency, which is correct.

However, as we move the ONEFlux standard behavior to not produce gapfilled values for long gaps (currently defined as gaps longer than 21 days), the calculation of _REF values need skip gapfilled data points for long gaps, so the final output of a _REF variable is computed using the same data points that are available for the individual percentiles.

Note that gapfilled values will still be saved in nee_proc and ure outputs, with long gaps being reverted to missing data in the final product generation step for ONEFlux (99_fluxnet step). The change is only for the calculation of the _REF variables, and not the contents of the input and output files.

[gilbertozp]

Additional info: to allow for experimentation with the lengths and types of gaps, the length and definition of gaps should be parameters for the functions. For instance, using the 21-day window proposed and QC flags larger than 0 define one way to filter/keep data points; but if the window size and QC value threshold are parameters, the code can be run with different values for evaluation.

It is likely the window size and QC value threshold will need to be changed in the future, which would also be covered if these are parameters to the execution.