Fix groupmod length calculation

[harmjan]

I apparently am doing something weird where I receive packets, edit them,and then send them out again. If a GroupMod already contains buckets will the length be incorrect after replacing them with new buckets.
I think this problem occurs in every function that contains "this->length_ +=", isn't a constructor and doesn't contain the string "add" in the function name. I'm not going to try and change all of them.
I also looked at the of13::FlowMod class since I also use that one a lot, the functions of13::FlowMod::match and of13::FlowMod::instructions technically have the same problem but the length of the of13::FlowMod packet is recalculated in the pack function so there is no benefit to changing them.

[ederlf]

I did not consider the possibility to replace objects from a class because I did not foresee message objects living after being sent or processed.

The pull request may solve the issue but it is just a quick workaround, so I will not merge as it is better to think about a different approach in the case of replacing members of different sizes in a message.

Thanks for pointing out the problem!