Update scimes.py

SCIMES function updated:
- cloudstering: the first part has been reordered. Moreover, the spectral clustering now works on a "reduced" affinity matrix, i.e. trunk branches made by two leaves only are recognized already as individual clusters and not considered in following the spectral clustering routine. This significantly increases the computation speed.

- guessk: the guessing of the cluster number has been made more efficient and reliable.

- mat_smooth: I am trying to use a local scaling approach. This generated a scaling parameter map that does not required a previous knowledge or estimation of sigma. Providing an user defined scaling is still possible, though.

- clust_cleaning: the assignment of clustered leaves to individual branches is more reliable now with a wiser usage of python sets. This should solve the problem of big cluster disappearing of the previous version.