Improved multi start solution selection

The logic used to determine the best solution from a multi-start run has been improved. The best solution is defined as the one with the highest simulator objective function, that is also consistent with the optimizer objective function. Using the simulator objective function ensures consistency in comparing different start points and eliminates issues due to optimizer tolerances and penalties.

Previously the two values were compared with an absolute tolerance, which in models with large objective function values could lead to the incorrect start point being identified as the best. The algorithm has been improved to use relative differences when comparing the solution value, leading to more robust identification, independent of the scale of the profit.