fix: Maximize value of all states in the LP program.

This does not affect the optimal value of the strategy, but makes the
reconstruction of the strategy easier.