Permit different array strides for fillsinks

fillsinks currently assumes that arrays are stored in column-major order: the dimension with size nrows should vary the fastest. This follows MATLAB, Fortran and Julia practice but the default NumPy array storage is column-major, and we need to copy NumPy arrays to a new storage order to ensure that the morphological reconstruction is done correctly. Column-major order is not a strict requirement of the morphological reconstruction algorithm, which only needs the strides of each dimension to compute the neighborhood of a given pixel. If we explicitly pass in the strides, we should be able to operate on row- or column-major data and on subsets or views of data without requiring copies.