Fix total elapsed time discrepancy between eventlogs and RTS

[Mikolaj]

Sometimes it causes a visible difference between the +RTS -s output and TS stats summary (GC times are exactly the same, but MUT is not, due to the difference in total time, and a few other stats that depend on MUT also suffer).

[Mikolaj]

TS total time is greater than reported by RTS, because we in TS don't subtract init and exit, which are usually below 0.01 seconds, but enough to affect rounding of total time.

We'd need an event for start_exit_elapsed and end_init_elapsed or a single shutdown event with total time. But that's still not enough, because in fact, the init, ext, gc and mut time in RTS don't add up, because the timing of their starts and ends is taken in different moments (e.g., start_exit_elapsed should be, but is not equal to end of tot_elapsed). I'd need to fix that first.

[Mikolaj]

10:53 @dcoutts mikolaj: if the time difference is really just the init/fini and is 0.01s then I expect we don't care 12:52 @mikolaj dcoutts: Closing. (to be reopened if ever users get puzzled by the tiny discrepancies)