Improve MPI-related and Simulation docs

[ziotom78]

This PR adds the following features to the documentation:

• It corrects the syntax used to turn MPI on/off in a Simulation object (we do no longer use the use_mpi flag);
• It adds a (long) docstring to Simulation.create_observations;
• It removes references to use_mpi that were still present in docstrings;
• It fixes a few minor errors in the Sphinx documentation.

[ziotom78]

This should fix issue #109. Note that I decided not to remove distribute, as I realized that there are significant use cases where it might be indispensable (e.g., when you do not want to rely on the standard way of distributing observations offered by the framework).

[dpole]

Thanks Maurizio, I am still a bit confused by all the possible interactions that can happen between the parameters num_of_obs_per_detector, distribute and n_blocks_time.

num_of_obs_per_detector chunks the time and keep the chunks as a list of observation. How the content in the memory of the process changes depending on the values of distribute or Simulation.mpi_comm?

[ziotom78]

I have renamed the confusingly-named distribute parameter to split_list_over_processes and have added a (quite long) docstring to Simulation.create_observations. @dpole , @paganol , this should answer the perplexities we had, what do you think?