Use unsynchronized buffered OutputStream for history

[SirYwell]

Overview

Description

BufferedOutputStream has a fair amount of overhead as it's thread safe. However, we don't require thread safety here, so we can use an alternative implementation from fastutil which performs better. In my tests, the time spent writing history is halved.

Submitter Checklist

• Make sure you are opening from a topic branch (/feature/fix/docs/ branch (right side)) and not your main branch.
• Ensure that the pull request title represents the desired changelog entry.
• New public fields and methods are annotated with @since TODO.
• I read and followed the contribution guidelines.

[dordsor21]

It's probably worth adding a note somewhere that it's changed to a non-thread-safe implementation

[SirYwell]

It's probably worth adding a note somewhere that it's changed to a non-thread-safe implementation

Where would you think is the best place for such comment? I was thinking about the processSet method in AbstractChangeSet as that's the point that synchronizes atm. However, the whole design isn't thread safe, writing to the stream from different threads would have caused issues even before...

[dordsor21]

Only because it's a public method in a util class, so probably a javadoc note on getCompressedOS

[SirYwell]

I saw that DataOutputStream#write is synchronized. Overriding and reimplementing that method brings even more improvements.

Overall, with send-before-history: false this brings down a //set stone on a region (0, 0, 0) -> (1023, 319, 1023) from >70s to <20s.

Because flamegraphs are cool, here a diff comparing the baseline to the changes on this branch (green: in baseline but not in patch, red: in patch but not in baseline. Unrelated calls are hidden)