Efficient Neighborhood Operations

When attempting to work with `ConvolveNaive` with a 1D kernel of size 3, the op throws `ArrayIndexOutOfBoundsException` right away because it attempts to position the input image at index `-1`. I believe this is correct behavior in the sense that the convolution is being computed at the image boundary, so an `OutOfBoundsStrategy` needs to be associated with the input image.

How should we address this? Should we update the `conforms` method to fail early if the given input does not have an `OutOfBoundsStrategy`? Or should we transparently extend the input with one in the case that it doesn't already have one? What do you think, @bnorthan? @dietzc?