s5bytestream: fix segfault (cleanup before trying next streamhost)
This segfault happened when none of the available streamhosts can be
connected to - or if at least one of them fails to connect.
Before this commit, it can be reproduced reliably by setting the "proxy"
setting of the account to nonsense, for example, this is what i used:
proxy.example.org,1.2.3.4,7777;proxy.example.com,173.194.42.65,80
jabber_bs_recv_handshake_abort() calls jabber_bs_recv_handshake(), which
is supposed to restart the handshake with the next streamhost. And it
replaced bt->tf->watch_out, which held an event ID, with a newer event
ID. So the replaced event ID doesn't get removed, and it gets called
again when its socket is closed by the timeout - and by the time that
happens, the memory is free()'d already. Boom.
The patch is simple - created jabber_bs_remove_events() to cleanup those
events, and use it before any code that expects to restart the cycle.
So basically the same as doing b_event_remove(bt->tf->watch_out).
I hope there aren't more bugs like this in this code.
