last_datapoint = invocation time for collectors

[hbilar]

We have a collector that occasionally takes ages to run (40-50 seconds), and sometimes a lot longer (600+ seconds).

Killing after 600 seconds is fine, but we found that tcollector.py kept killing this collector in check_children() for newly spawned collectors, as the last_datapoint was too old on every invocation of check_children. Also, because it was being killed off all the time, the last_datapoint obviously never got updated after it hit the 600s limit the first time.

The fix around this (well, a fix that works for us, at least) is to set the last_datapoint of a collector to the time it was spawned, rather than when the object was first created (which could be ages ago).