perf: Optimize Bitmap Iter

[kwannoel]

#7792 (comment)

1. Instead of using iterators, just loop over chunk + bitmap directly. Iterators have at least 1 jmp for checking if there's any items left. So with 2 iterators we have 2 jmp (if not optimized away). See if this improves things.
2. Bench with and without caching step for bitmap iter, see how things change for q17 benchmark / streaming fold agg.

Let me know if any further details are required, I can definitely provide more elaboration.

[kwannoel]

Here's the baseline that is expected: #9129. Around a 5% improvement in the hashagg q17 bench when we do compact.

If BitmapIter optimization does better, we should definitely use it instead.

[kwannoel]

This can be closed, bitmapiter no longer shows up in flamegraph.