Add tutorial comparing numerical optimization with MCMC

[HDembinski]

The tutorial should compare optimization of a likelihood with a parametric pdf with a MCMC generation of the posterior.

Numerical optimization should be superior to MCMC when the model is a parametric pdf and there are few parameters to optimize. MCMC may or may not perform better in high-dimensional parameter spaces.

MCMC is clearly superior when no parametric pdf is available for the model. With MCMC, one can infer the posterior pdf of the parameters even if the model is some kind of Monte-Carlo generator.