Repeated computation of co-occurrence matrix in haralick features

Hi,

A question on haralick features.

All haralick features call the `getCooccurenceMatrix` function on the input image. If I want to compute N different haralick features on an image, the same co-occurrence matrix is computed N times. Unfortunately, the computation of this matrix is often the most intensive part of the feature computation. Ideally, this matrix would be precomputed once per image.

Perhaps this can be solved by overloading the `compute` function, adding a signature that takes a `final double[][] matrix`? I am not sure in what interface or class this signature should be defined, however. Or maybe there is a better way :slightly_smiling_face:

Start by locating the Haralick feature implementations and the getCooccurenceMatrix entry point, then trace how compute is exposed across the relevant interfaces or classes. Define the reuse boundary so one image's co-occurrence matrix is not recomputed for each feature; done means the API supports this without changing feature results and coverage verifies the behavior.

Written by the indexing model from the issue text.