Unbalanced add router: settle at the end

[jubeira]

The implementation in #1423 is simple and works, but it settles each step separately (first add, then swap).

The problem is that the sender actually be doing a single token add, without necessarily holding both of the tokens (which is in fact what we do to test the code). This will revert in case the holder doesn't have both of the tokens.

We can do the operations in a transient manner, and settle at the end to avoid this issue.