SCRUM-5949: Fix facet sort, routing distribution, reduce BP threads

[cmpich]

Summary

• Fix category facet sort order to sort by count descending
• Fix variant indexer shard routing: use large random int instead of 0-15 range to ensure even distribution across all shards (was leaving 5 of 16 shards empty)
• Reduce RoutedBulkIndexer threads from shardCount4 to shardCount2 (32 per species instead of 64) to reduce ES client count and memory pressure
• Add ES client identity logging to diagnose unclosed client leak that prevents JVM exit

Test plan

• Verified empty shards caused by limited routing range on stage cluster
• Verify even shard distribution after next variant indexer run
• Verify client create/close logs match (no leaked clients)
• Verify JVM exits cleanly after indexing + snapshot completes

[claude]

Code Review

The change looks correct. It replaces a hardcoded category display order with a simple descending sort by count (Long.compare(b.getTotal(), a.getTotal())), which matches the ticket intent (SCRUM-5949).

No bugs, no performance concerns, no security issues. The removed CATEGORY_ORDER list and associated index-lookup logic are cleanly eliminated. The sort comparator correctly uses Long.compare to avoid overflow pitfalls of subtraction-based comparison.

LGTM.