Improve performance of arrayDiff

[mhahnenberg]

Is your feature request related to a problem? Please describe.
We've observed arrayDiff in some profiling we've done in production when subscribing to a topic with a nontrivial number of partitions (e.g. > 1000) with a relatively high fetch rate. After checking out the code it appears that arrayDiff is written in a way that is O(n*m) since it's iterating over a and calling b.indexOf() for each element.

Describe the solution you'd like
Using a Set could be an easy way to make each of the checks inside the loop O(1) instead of O(n) with indexOf, thus making the overall function O(n+m) rather than O(n*m).

Additional context
I've written a simple PR that does this, will attach after filing this issue.

[mhahnenberg]

PR for a simple fix is #1738