Add alpha, cell_alpha and b_tau for npt_langevin_step in NPTLangevinState

[thomasloux]

A recent change of API #264 as modified the way one uses an integrator. Before one would invoke npt_langevin that would provide a init_function and step_function with relevant arguments kept fixed during the simulations, now one directly uses npt_langevin_init and npt_langevin_step. Yet, in the case of Langevin NPT, this integrator needs 3 parameters: alpha, cell_alpha and b_tau.

These arguments can have default values in npt_langevin_init but not yet in npt_langevin_step. That is a bit annoying in the case one just want to use default values. So I would suggest to put these variables in the NPTLangevinState as NPTNoseHooverState that also has a b_tau and t_tau variables that are actually stored in NoseHooverChain object. alpha, cell_alpha and b_tau should be system attributes.