turbulence models with extend_simul_class

This is a quite foundational MR which fixes #116 (closed) (Simple LES and RANS models added with extend_simul_class).

There are quite a lot of changes at different levels:

• Base fluidsim things (in fluidsim/base/solvers, fluidsim/base/forcing, fluidsim/base/turb_model/base.py). Just interesting for fluidsim core developers.
• Implementation of a model (class SmagorinskyModel in fluidsim/base/turb_model/smagorinsky.py)
• Testing (fluidsim/base/turb_model/test_turb_model.py)
• Examples (doc/examples/kolmo_flow and doc/test_cases/Taylor_Green_vortices/run_simul_les_smago.py)

If you want to take inspiration to implement another model, you should look only at fluidsim.base.turb_model.smagorinsky.SmagorinskyModel and the associated test class.

More work needed on this subject:

• For now, it works only for the ns3d solver
• The outputs do not take into account the forcing added by the turbulent model (see for example sim.output.spatial_means.plot_dt_E)
• It would be nice to improve testing by checking that the model works as expected
• It would be nice to implement the dynamic Smagorinsky model

A question: for now the model is recomputed each time turb_model.get_forcing is called, i.e. each time sim.tendencies_nonlin is called (for example 4 times per time step for the RK4 time stepping scheme). Would it be sufficient to compute it once per time step?

Warning: to try this code, one must really reinstall Fluidsim from source, which means update the source (hg pull; hg up turb_model), and then cd lib; pip install -e .; cd .. (to install the modified version of fluidsim_core) and then pip install -e . (to recompile a modified extension).

One can check that everything is fine by running pytest fluidsim/base/turb_model.