Stops using IndexSearcher::count in Lucene.

[archolewa]

--It looks like IndexSearcher::count fires off an entire query, and
then counts the number of documents that were hit. Currently, we are
using it to determine the total number of dimension rows that satisfy a
given query. However, this is unnecessary and wasteful, because the
TopDocs object returned by the query for the actual data also
contains information on the total number of documents hit by the query.

[michael-mclawhorn]

I would put a CHANGELOG entry around this even though it's a small internal change.

[michael-mclawhorn]

We don't actually test the internal behavior but I can probably live with that. Once I've checked out the changelog I should be able to approve this.

[cdeszaq]

👍 And I agree w/ Mike, it would be good to call out this expected performance boost in the Changelog (50% fewer Lucene queries in the typical case of getting the 1st page)

[archolewa]

@michael-mclawhorn Is the test I added insufficient somehow? In our case, we only use the documentCount to populate the pagination metadata, and I verify that the pagination metadata is correct.

[cdeszaq]

I think the test this PR adds is a good addition. I think Mike's point is more that we're making an internal code change which should result in a performance boost, but we're not really checking the internal implementation anywhere.

Personally, I'm 100% OK with not testing this particular difference:

1. It doesn't (shouldn't) change the correctness of the result
2. It's going to be difficult to test, since it's very intimate with the class itself (and even that particular method)
3. I don't think we get much value from testing it