Ordering of cells in CubicalComplex

[mglisse]

@MathieuCarriere 's new function cofaces_of_persistence_pairs returns an index into a flattened version of the array of top dimensional cells. While the constructor taking a numpy array had somehow hidden the unintuitive ordering of the cells, this exposes it again.
We need to add some description of the order. When adding the constructor from numpy array, we had a choice between 2 versions (IIRC: fortran order, or reverse the dimensions) and picked one randomly because it didn't matter. Well, now it matters again, so we need to decide which one we want.
We could also provide the output in a different form (with some option?). numpy.nonzero for instance does not return an array of indices into a flattened array, it returns d arrays corresponding to the indices in the d-dimensional array. This form of output has the advantage that it does not depend on the internal ordering, and it is ready for indexing.