Convenience methods for signed distance transform

[hanslovsky]

With the newly introduced binary distance transform methods (#69, #71), we can now very easily create a signed distance transform (beware: Kotlin!):

private fun <B: BooleanType<B>, T: RealType<T>> signedDistanceTransform(
		mask: RandomAccessibleInterval<B>,
		distanceOutside: RandomAccessibleInterval<T>,
		distanceInside: RandomAccessibleInterval<T>,
		vararg weights: Double = doubleArrayOf(1.0),
		distanceType: DistanceTransform.DISTANCE_TYPE = DistanceTransform.DISTANCE_TYPE.EUCLIDIAN
): RandomAccessibleInterval<T> {
	DistanceTransform.binaryTransform(mask, distanceOutside, distanceType, *weights)
	DistanceTransform.binaryTransform(not(mask), distanceInside, distanceType, *weights)
	val paired= Views.interval(Views.pair(distanceOutside, distanceInside), mask)
	return difference(paired)

}

with helper methods

fun <B: BooleanType<B>> not(mask: RandomAccessibleInterval<B>): RandomAccessibleInterval<B>
{
	return Converters.convert(mask, { s,t -> t.set(!s.get()) }, Util.getTypeFromInterval(mask).createVariable())!!
}

fun<T: RealType<T>> difference(pairs: RandomAccessibleInterval<Pair<T, T>>): RandomAccessibleInterval<T> {
	return Converters.convert(
			pairs,
			{ s,t -> t.set(s.a); t.sub(s.b) },
			Util.getTypeFromInterval(pairs).a.createVariable())!!
}

This will be negative inside the object defined by mask and positive everywhere else.

[axtimwalde]

Happy to add this once we decided what to do about the fact that this steps from -1 to 1 skipping 0 per definition of the distance transform boundaries being in the boundary pixel and not between boundary pixels (lab discussion, @hanslovsky knows the details). In isotropic spaces, we could probably just subtract 0.5 from all distances which states that the boundary is 0.5px closer to the pixel without making an effort to explain where it is. But this would not work with non-isotropic grids. This means this offset should probably go into the DT algorithm?