More accurate interpolation

[Rylern]

Considering a point p and its four closest points p1, p2, p3, and p4, linear interpolation is implemented with the following formula:

p = t1*p1 + t2*p2 + t3*p3 + t4*p4

with the ti between 0 and 1 representing the distance between p and pi.

With integer images, the ImgLib2 NLinearInterpolatorFactory computes something like this:

p = round(t1*p1) + round(t2*p2) + round(t3*p3) + round(t4*p4)

This is an issue, because there is a high loss of precision. For example, with an image containing only pixels with value 20, this interpolation can create pixels with values 19 or 21.

This PR provides a custom implementation of the linear interpolation, which computes something like this:

p = round(t1*p1 + t2*p2 + t3*p3 + t4*p4)

There's still a precision loss, but less than before (and it's unavoidable).