This removes the g_PlayerbaseTypes global from rmgen2, as this prevents adoption from playerPlacementByPattern() by rmgen1 maps (since they would need to depend on rmgen2 for that).
Since the information if walls can safely be placed is still necessary, these values are moved inside playerPlacementByPattern().
This is an easier solution than D5082 (which solves the same problem, but more complicated and possibly also more unfair).