Revert the "q.alt" performance "improvement"

After doubts came up about whether our q.alt performance improvement (introduced in #1773440: Use q.alt instead of fq for “filter only” queries) actually constitutes an improvement, not a detoriation, I asked on the solr-users mailing list and got this response.

Which does make sense, of course, and since nobody else replied (so far), it seems that's about it.
If someone wants to still run some performance tests, that would be really great, but otherwise I think we're just going to drop this change again. (An additional problem with using q.alt is that it influences relevance – cf. #2466705: Weird relevance when searching with filters only (no text search).)

Especially in our module, where every single search contains the same two filters on hash and index_id (though the latter can of course have a few different values), it seems insane to disable the Solr filter cache for those.