Currently the split groups are not deterministic when adding or removing (or even moving and existing test to a different location in the SConscript). The tests are simply round robin added to a group to make sure the groups are evenly distributed.

This causes test to move groups and is confusing for developers and the auto reverter.

Instead we should hash the path of the test binary and create groups based on the first letter of the hash str. This will provide consistent groups while also creating mostly even distribution (based off random chance).

The pool of characters is [a-f0-9] or 16 possible characters. The groupings would be

group 1 [a-d] (4 chars)
group 2 [e-0] (4 chars)
group 3 [1-4] (4 chars)
group 4 [5-9] (4 chars)