Hash based identifiers rather than counters for tests and runs

[symbiont-stevan-andjelkovic]

Monotonically increasing counters are currently being used as identifiers for tests and runs of tests, using hashes instead would give us several advantages:

1. Avoid rerunning the same test twice (given that the tests should be deterministic...);
2. Share tests between users/CI by hash, thus avoiding clashes in the db (i.e. the same run has different counter ids);
3. Concurrent runs would less likely cause identifier races (if we are not careful two threads could still run the same test and therefore create a hash clash).

The disadvantage of using hash identifier is they are longer.

What exactly to use to create the hash remains to be figured out. It should include everything that makes a run unique and reproducible, i.e. the test, seed, faults, git commits of all software involved, etc.