bolt: Improve delete inactive orders/matches perf.

[jholdstock]

Running these functions on a resource contrained machine and a large database was incredibly slow due to overuse of the newestBuckets func. Every key was accessed and sorted every time a batch of orders/matches was processed.

Rather than repeatedly performing this work, the full set of keys is now accessed only once. The resulting code runs ~200x faster.

There is a slight change in behaviour because database transactions are now committed with a maximum of 1000 deletions rather than exactly 1000. This is not an issue because the exact size of each transaction doesn't matter, the important point is to limit the maximum size of each transaction to prevent excess memory consumption.

This performance gain removes the need to log the progress of each individual batch, so now only a summary is logged after the entire deletion process is completed.

[jholdstock]

Just reviewed history and noticed @JoeGruffins was the original author of this code. Care to give it a once over Joe?

[JoeGruffins]

I can't remember writing this but changes look good and tests passing.