Maboss

Two bugs in the parser that Gautier found :

- Refstate should allow -1 values (which ignores it)
- A node can have non-defined rates. In this case, default rates are used

Added a first version of the MaBoSS C library interface.
For classic simulation, and upmaboss simulations, setting cmaboss=True in the run() command will use the C library.

Also added a option to run through a optimized engine, which only returns the probability of the last states. For classic and upmaboss simulations, setting only_final_state=True in the run() command will use this new engine.

When computing the node probability trajectories, we would only return a node probability when it was non-zero. Now we uses the defined outputs of the model, ensuring all outputs would be present even if their probability is zero.

I also added a direct maboss.loadBNet(file) function, to load a bnet file as a maboss model directly (without explicitely callling biolqm).

Finally, when selecting output variables, there was no error checking that each variable was indeed a defined node. Now the library will return a failed assert if it is the case.

There fix removes biolqm dependency, by loading the model of the ensemble using the bnet parser in minibn.
This also allows to have the formula in the proper, simplified form

- Mostly ensemble modelling code (documentation to come)
- Bugfix for simulation palettes