Add power exponent to the interpolation distance

[GieziJo]

knn_interpolate assumes inverse square distance as contributing weights for the interpolation.
In certain cases, you might need linear interpolation, or another power.

I added the parameter alpha to the function with the default value set to 2.0.
The default behavior thus stays the same, but gives you the option to change the power.

To keep the performance the same, added a couple of checks:

        if alpha != 2.0:
            distance = distance.sqrt()
        if (alpha != 2.0) and (alpha != 1.0):
            distance = distance.pow(alpha)

Hope this can be useful!