If an MLCP solver fails, use PGS/SI fallback, add a boolean return value for 'solve' method
Added a PGS solver that uses the sparsity of the A matrix, just for testing (the equivalent sequential impulse solver is much faster, not having to allocate the big matrices)