Commit Log Write Performance Improvements

[prateek]

We have talked about performance improvements for the commit log a few times, going to start tracking the different experiments/ideas in this ticket.

---

• Idea: batched write APIs Add BatchedWrite methods in the Database constructs #458

---

• Experiment: Rolled out a version of the commit log code which does not make any writes to the disk, but does everything else for commit log processing ([DO NOT LAND] Drop all commit log writes #944) to a single production host. Here's what it looked like v another host. The host with the /dev/null build is in green.
  With the caveat this is a rudimentary experiment, we still do observe high spikes in the commit log queue size.

/cc @richardartoul @robskillington

[richardartoul]

@prateek I'm glad that the experiment showed it wasn't a disk issue. I have a few potential fixes I'm gonna test out today and then document the results.

[richardartoul]

I feel like this ticket was mostly addressed by #1157 and #1160

The one last thing I'd like to do is making it so that when you open a new commitlog file, it actually just lets you immediately start using one thats already ready and then it asynchronously creates a new one so that file creation doesn't block the queue for any amount of time during rotation

[richardartoul]

All of these tasks have been completed