Large result sets in distributed settings are slow

[tokee]

Poor decision logic in the code means that while phase 1 of distributed faceting is processed with the optimum code path, phase 2 might not be (if there is no cached counter available).

The logic of phase 2 should be

• If a cached & filled counter is available, use that for fine counting
• If no previously filles counters are available, check the result set size. If the size is above a given threshold, switch to vanilla Solr counting. If it is below, continue with full counting.

[tokee]

Fixed by commit 392e74648d091425358764aeb1fbd6edc15a3733