bridge connection failure delayed by keepalive_interval

[abiliojr]

This bug happens using ADNS with a bridge.

Any bridge connection attempt that fails more than once seems to be delayed by keepalive_interval, and not only by restart_timeout.

Log sample, using restart_timeout = 3 and keepalive_interval = 5:

1541844877: Connecting bridge (step 1) amb (localhost:1884)
1541844877: Connecting bridge (step 2) amb (localhost:1884)
1541844877: Socket error on client local.abiliojr.amb, disconnecting.
1541844881: Connecting bridge (step 1) amb (localhost:1884)
1541844881: Connecting bridge (step 2) amb (localhost:1884)
1541844886: Socket error on client local.abiliojr.amb, disconnecting.

See that in the second attempt, 5 seconds pass until the socket error is detected.

This seems to be caused by a reuse of the context without proper cleaning, leaving context->events dirty. This makes the condition if(!(context->events & EPOLLOUT)) in file loop.c to be evaluated to false in the subsequent attempts, changing the behavior and triggering the extra delay.

[abiliojr]

Solution in pull request 1029