Don't republish call capability on forward failure
When forwarding of a call fails for some reason, the kernel
automatically answers the forwarded call with EFORWARD. It must not
republish the capability because only unanswered calls are associated
with a capability (so that they can be forwarded or answered).
Republishing the capability makes it possible for userspace to attempt
to answer such a call again, which results in kernel panic due to a
failed assertion. As an interesting fact, uspace functions
async_data_read_forward_fast and async_data_write_forward_fast do this.
Also, the motivation behind republishing the capability as explained
by the removed comment turns out to be invalid because such a call
ends up among the original caller's answers.
