Layering returns different results for the same dataset

[fpavogt]

Describe the bug
ncom_from_gmm() was found to return a different number of best_ncomp for certain datasets, if run multiple times.

This was related to the lack of fixed random seeds in the code, which leads to different results for "borderline" datasets: specifically, very thin groups that have a nearly quantized distribution of hits.

Expected behavior
The code should always return the same number of layer components.

[fpavogt]

Fixed by specifying a random seed in the code (optionally set by the user).