Revert "Optimize DocIdSetIteratorAcceptDocs#cost (#15592)"

[benwtrent]

@shubhamvishu sorry for the revert :(. I ran into some weird test failures and it turns out (unsurprisingly) mutating the behavior of cost depending on if some other method was called is a pretty rough API.

For now, I am going to revert, and I will likely have to rebuild another RC for 10.4.

I think if we are going to do this, we need a cost and an approximateCost API. Where approximateCost can change, but cost is when we want the real cost :(

[benwtrent]

The specific failure is that when we do cost checking in the HNSW format, sometimes we check before the idset is created and other times after. Which pretty much breaks downstream search logic that depends on accurate costs.

[shubhamvishu]

Thanks @benwtrent for identification and reverting the change!