Use Vanilla Java instead of Stream API for performance critical sections

As benchmarked by @mad (https://github.com/JanusGraph/janusgraph/pull/2098#discussion_r415074443) and @rngcntr (https://github.com/JanusGraph/janusgraph/pull/2105#discussion_r419973555) Vanilla Java has better performance characteristics then Stream API.
Thus, I am suggesting to use Vanilla Java code in performance critical sections.

Start with the benchmark evidence in PR 2098 discussion_r415074443 and PR 2105 discussion_r419973555, then locate the performance-critical sections using the Stream API. The issue does not name files, entry points, or benchmarks, so the scope and completion criteria need to be established before work begins.

Written by the indexing model from the issue text.