Adjust a formula to ignore hidden/filtered rows of data

I wonder how this can be improved on large datasets (e.g., tens or hundreds of thousands rows). VSTACK, basically, replicates *all* visible elements of a column, which leads to higher memory consumption. On the other hand, using MATCH to identify whether we have the current element 'v' in the accumulator variable 'a' will lead to poor performance due to search being executed on an unordered dataset thousands of times...