Replace destroy in the failed SA with DESTROY_IF

[TNLTom]

On testing, we get random segfaults on this particular destroy call. I can't really reproduce it but replacing the destroy call with DESTROY_IF should not hurt and since then we have not seen the segfault anymore.

[tobiasbrunner]

You should investigate when this happens exactly, because there should not be a call to this function with this->child_sa being NULL. It's called from process_i* implementations, that is, after build_i, which always allocates this->child_sa (and that can't have failed as there would have been a crash earlier). So there has to be some other issue here. What failure triggers the call to delete_failed_sa()?

Also note that the ikev2-qske-multi-ke branch will not be developed any further, the current branch with multi-KE support is six-beta.