Improvements to the algorithms

I've improved the Algorithm for the palette interpolator slightly. By "cleaning up" the inputs, much better results can be returned by the found regressions. Without going into the details too deep, I remove consequitive values from the input if they are one of the two extremes (0 or 255). Without those values, the regression doesn't try to hit those points, leading to less overfitting. Simpler regressions with a more natural shape.
To illustrate my point, take a look at this graph I have made. This shows just the blue channel from TestPalette.png (included in the tool). The dotted line is the exact colours, the thin line is the found regression function using all the points. The thick line is the functions with the inputs cleaned, which means 1 point was removed. The difference is obvious to the naked eye. Because of the point (0.2, 0), the old function ackwardly weaves itself around 0, meaning from 0.5 to 0.67, it returns 0 when it probably shouldn't. The new function however is above 0 from 0.5 onwards, leading to (little to) no blue component when there shouldn't be and some when there should. Simply said, it's just better. Download it and give it a go.