Convert RQA result to UInt8 immidiately

[felixcremer]

We can save on a bit of disk space and also intermediate allocations if we discretize the rqa results directly into UInt8 during the computation. We would have to set the range that we would like to represent in UInt8 from the actual data we already computed.
To do that we would like to look at histograms of actual change detection data and then use an offset and scaling factor to compute descrete classes between 0 and 255.

We would then also swap the rqa results and have the no change class being 0 and the highest confidence in change at 255.

This would also make the interpretation of the results easier.

[felixcremer]

This was closed by #101.