Enable write-ahead log, recover in-progress TXes at startup.

[mystenmark]

A few items are still TODO:

• Tests (coming soon, want to make sure there's consensus on the approach before I work on this)
• Offline retries - right now we only process the log at startup. The main motivation is for crash recovery, so that's ok. But we'd also like to attempt recovery periodically while running in case txes fail transiently due to db write errors.
• After tests are done I will:
  • Remove object critical section.
  • Break up the big atomic batch write for better performance.

[gdanezis]

Nice one -- I have a few questions around the "pop" in the recovery logic; the conditions under which we might retry; as well as deleting from the retry table.

[mystenmark]

@gdanezis thanks for the excellent comments, will address the ones that require edits!

[velvia]

Left lots of comments.... good work @mystenmark