Refactor bankstats reading in finalsink.

In preparation to dump/read all bankstats files, I've refactored the finalsink code to read/combine bankstats for calcfap and combine_stats.

Tests

I've tested on a short bg run, an 8000s inj run, and a bg run with short snapshots.

The bg runs before & after the refactor have identical bankstats, but different marginalized stats. I think that's just due to when the bankstats dumps/calcfap processes started/ended. As a consequence, the zerolags have different FARs, same as when you do 2 bg runs before this change.

The injection run is identical before/after: test_refactor.log** (bankstats are also unchanged)