Optimize replicateA to have same recursive structure as replicateA_

[johnynek]

follow up to #4208

This implements replicateA in the same was as replicateA_. The benefit over using traverse, which was the previous implementation, is that traverse ignores that all the items in the sequence are identical. This implementation leverages that to get O(log n) calls to map2 instead of O(n) calls to map2.

Note, it is still an O(n) method in the worst case since finally we have to materialize the List[A] which is length n, but often it is the map2 calls that are expensive, not iterating the list.

[armanbilge]

This is really neat! What do you think about a tail-recursive implementation? I gave it a try in 1b55564.

[johnynek]

I personally don't think it is worth it because the depth is already log N. So even if you use Int.MaxValue here you are dealing with 31 recursive calls.

If you're really interested, we can do it. A loop is better than depth 31, but I think the trade-off is more complex implementation in return for a very rare case where you can afford 1 additional call but not 31.

[armanbilge]

We can merge yours and I might put up mine as another PR 👍