Optimizations to golr loader, removes MapDB

[kshefchek]

The golr loader currently takes 30 hours to load all documents excluding optimizing, moving files, setup/teardown on the 201902 release of scigraph, using the queries on https://github.com/monarch-initiative/monarch-cypher-queries/tree/60a3cf

This update improves the load time to 16 and a half hours with the following updates:

• Replaces all UNION queries with COLLECT and ordering the output by subject, object
• Removes the need for MapDB to store all neo4j results for each query and then sorting, instead receives the sorted data and operates on it accordingly
• Increase batch commit size from 1k to 10k docs
• Decrease n threads from all avail processors to 7
• Increase java heap max on golr loader from 90G to 300G
• Increase neo4j pagecache from 50G to 100G
• Increase Jetty idle timeout from 50 seconds to 200
• Increase solr heap max from 4G to 10G

The downside is that we monopolize all the memory on one of our servers for ~8 hours.

See the accompanying PRs: https://github.com/monarch-initiative/monarch-cypher-queries/pull/33 https://github.com/monarch-initiative/solr-docker-monarch-golr/pull/1