Allow passing in OCL arrays to affine transform

[jni]

This gives a massive speedup when repeatedly applying affine transformations to the same volume, as in the case of image registration. (See this commit on the skimage affine registration PR, which requires these changes.)