Add rf_clip and rf_rescale and rf_normalize

[metasim]

@vpipkt Interested in your thoughts on this.

[vpipkt]

I have looked into such things a little in the past but never gone very far. Seems maybe geotrellis IfCell could be useful and may be we can expose something like that along the way.

rf_rescale is interesting because that can be done on a per tile basis, but could also imagine a case where the user wants to consider the column min and max. See optional parameters. Similar sentinment for normalize. Easy to do it locally, more powerful to consider entire column mean and standard deviation, but less clear about implementation.

I think there may be some functionality for rescaling in geotrellis, but not exactly sure.

Possible function signatures / doc

rf_ifcell

 Tile rf_ifcell(test_tile, true_tile, false_tile)
 Tile rf_ifcell(test_tile, true_scalar, false_scalar) 

(possibly with (tile, tile, scalar) and (tile, scalar, tile) sigantures also

Return a tile full of true_tile or true_scalar values where test_tile is true, else false_tile or false_scalar. In evaluating test_tile non-zero data values are true, zeros and NoData are interpreted as false.

An alternate name may be rf_where (after numpy).

rf_clip

 Tile rf_clip(tile, min, max)

Return a tile where values below min are set to min and values above max are set to max.

rf_rescale

Tile rf_rescale(Tile tile, [Double min, Double max] )

Rescale cell values such that the minimum is zero and the maximum is one. Other values will be linearly interpolated into the range.

If specified the min parameter will become the zero value and the max parameter will become the max. Values outside that range will be clipped to 0 or 1.

rf_standardize

rf_standardize(Tile tile, [Double mean, Double stddev])

Standardize cell values such that the mean is zero and the standard deviation is one. If specified the mean and stddev are applied to all tiles in the column. If not specified, each tile will be standardized according to the statistics of its cell values.

[metasim]

One thing to work out is that in the case where you don't provide the scaling parameters and expect the system to compute it for you, you can't actually have a columnar function (a regular expression) invoke an aggregation first. Either the user would have to compute the summary statistics first and pass the resultant structure in, or pass in the dataframe for the code to invoke it first.