Replace fill values by NAs in landsat pre-processing functions

[loicdtx]

Not sure whether this is new or not, but ESPA data include a fill value (20000), which I've seen appear in some RasterBricks processed by bfastSpatial recently. See product guide. This fill value is not considered by any preprocessing functions at the moment. Should be replaced by NA as early as possible in the processing chain.

[GreatEmerald]

That's the "saturate value" (according to the XML metadata of ESPA-provided surface reflectance-derived VIs). The "fill value" is -9999. Indeed, both should be filtered out.