Minor improvements to clarify semantics of event timestamp

While preparing for the XDC talk, I found that a few comments on event timestamps were not aligned anymore with what we found out about its actual meaning. This tiny MR tries to fix that. In the atomic event handler, I also renamed completion to event_time, since time of completion of the commit, and the actual meaning of the event timestamp are not always correlating (i.e. for drivers with high-prec time-stamping capabilities).

Also tossing in a small workaround for the compiler to not complain about unused vars.