One of the recent port forwarding crash reports contained details

One of the recent port forwarding crash reports contained details
which suggested bufchain_prefix() was finding an improperly
initialised bufchain structure. Looking at the code, this may indeed
have been able to happen, since the bufchain in a SOCKDATA_DORMANT
channel was not initialised until CHANNEL_OPEN_CONFIRMATION was
received. This seems utterly daft, so I now call bufchain_init()
when the channel structure is actually created. With any luck the
crash will mystically disappear now (I wasn't able to reproduce it
myself).

[originally from svn r1735]