The correct truncation is 128-bit but some implementations insist on using 96-bit truncation. With strongSwan this can be negotiated using an algorithm identifier from a private range. But this doesn't work with third-party implementations. This adds an option to use 96-bit truncation even if the official identifier is used.