Automatic init

Default init is 0.1 for all variables.
We could use the information from the OCP to improve the initial guess:
- variables with both lower and upper bounds (e.g. control) could use median value.
- back to the ODE aspect, state components with both initial and final conditions could use a linear initialisation; components with only one condition could use constant initialisation.
- trick for variable time problems (probably too hard to automate, could be given as a hint for users): if possible, use dynamics and boundary conditions to estimate the final time. Simple example: component x with derivative v, with boundary conditions x_0, v_0 and x_f. Variable final time could be initialized with (x_f - x_0) / v_0.