Avoid identical computation in self tanimoto similarity

[JochenSiegWork]

The ´self_tanimoto_similarity´ function equates matrix_a to itself, and then it calls the tanimoto_similarity_sparse. Calculating norm_2 is repeated in this case which is unnecessarily costly for large arrays. See

MolPipeline/molpipeline/utils/kernel.py

Line 29 in 8190785

norm_2 = np.array(matrix_b.multiply(matrix_b).sum(axis=1))

We can add a simple check for identity of the two matrices to avoid redundant computation.

Thanks to Afnan for bringing this to our attention!