Fix Parameter Generator

The generation of parameters for the Lorenz example was presenting issues:

• Parameters are always the same, the iter_param_set was not actually iterating;
• Active learning, which requires on the fly change of the parameter was not easily feasible.

This MR introduces a ParameterGenerator class that exposes a draw_parameters function that can work with iterators and be changed on-the-fly for active learning. The server only call the draw_parameters that can be directly instanciated as for HeatPDE or interfaced with a ParameterGenerator class as done now in the Lorenz example.