Log upload latency

In !151 (merged) we fixed a latency issue by making gracedb uploads async, but replaced end-to-end latency with pipeline latency in latency_history.txt

This change records 'end-to-end' latency, but adds timers & additional fields for each upload (coinc & log)

Along the way I did a few refactors, with a few deliberate side effect:

• Changed the order of some processing of 'significant triggers'. We should now write to disk (coinc & trigger_control) after launching the upload thread instead of before.
• There's a variable in finalsink self.last_submitted_trigger which was used in a strange way before. It should be cleaned up, I don't Think there's any functional change there. Maybe worth confirming.
• The 'coinc document' is now created before any given upload for code neatness. I doubt that the initialization is very expensive, if it is, we can go back to doing it at the end of the upload as we were before this change.

Tests

I'm running an 8000 inj test on ozstar with atomic add. I'm also creating 'latency_history.txt' by adding export GSTLAL_LL_JOB=$i to the pipeline script ($i is the bank, 13, 14, 19 in my tests), and reading it with gstlal_periodic_get_urls 13 14 19

I'll check logs for trigger_control.txt, stderr, and latency_history to check these changes.

It's also running in MDC.