return error estimates for CBN, HESBCN, etc

[rdiaz02]

Those are needed to get the likelihood and compute cross-validated likelihoods, etc. Right now, the predicted genotypes are those for models without noise.

• H-ESBCN returns the best epsilon (see function markov.h)
• CBN: see line 286 in h-cbn.c : write the epsilon and lik. to a file.