Extracted timestamps are wrong

[sanchda]

libdatadog specifies that sample timestamps mark the end of the event. Ours mark the beginning. This should probably be fixed, although I'm not 100% sure how to do this in a general way.

• For timelike events, this is clear
• For events in which durations must be extrapolated, this is slightly unclear for lack of examples.

Not really fixing because it doesn't seem to me that there's a big difference between estimating our CPU time stimates as forward or backward durations (I think????)