If downstream connection pool crashed and recovered while some
upstream was alive holding it's DC ID, it may return downstream
taken from the old incarnation of pool to new one because it's
registered under the same name.
Fix is to hold pool's PID instead of name

By checking for monitor message

* Trigger GC with lower threshold
* Unconditionally die if queue len is too large

There was a race-condition, that caused crashes.
We could use gproc, but it feels a bit overkill for ~10-20 processes

use fully qualified node name