[FEAT] Implement payoff gradient function in MixedStrategyProfile #1131

In many methods (including those involving path following in some form or another), the derivative of the expected payoff of a strategy with respect to all other strategies is needed. This can be computed efficiently in a single pass by re-using computations. This is not exposed publicly in MixedStrategyProfile.

Add a method

virtual void GetPayoffDerivRow(const GameStrategy &p_strategy, Vector<T> &p_deriv)

which returns the derivative of the expected payoff to p_strategy with respect to all strategies in the game.

• For action graph games, this is basically already AGG::getPayoffJacobianRow.
• For table games, this can be computed efficiently by making a single ProductDistribution for each (other) player and computing the payoffs against that distribution.
• The base class should provide a fallback which just calls the existing derivative

This can be deployed in logit, hp, and leap to replace existing hand-rolled Jacobian rows.