Return value on error makes not much sense for sensitivity analysis

[FWuellhorst]

If simulations fail during a calibration, an option to return a pre-defined number (e.g. inf) makes sense. For an SA however, this makes little to no sense. Moreover, a failed simulation results in errors when passing the scalar return value to the functions of the SALib.