investigate: any faster ways of get_hash_values

[fuyufjh]

A few alternative approaches to make it more "vectorized" which I think worth taking a try:

1. Replace visibility Bitmap with visibility Vec<usize>. This is because for idx in vec is more vectorized i.e. no branches at all so more friendly to CPU pipeline.
2. Similar to 1 but only do it for low-selectivity cases
3. Do compaction before exchange, because we know exchange will always produce low-selectivity results.

Originally posted by @fuyufjh in #15696 (comment)

[github-actions]

This issue has been open for 60 days with no activity. Could you please update the status? Feel free to continue discussion or close as not planned.