Split T900_distributed.py module into separate sub-tests with self-test.py refactor

[klimaj]

This PR accomplishes several things:

1. Splits the PyRosettaCluster unit tests specified in the T900_distributed module into separate modules (T901–T921).
2. Splits the PyRosettaCluster unit test modules into separate modules (as discussed in PR Improving resilience of PyRosettaCluster upon Dask worker preemption #630)
3. Provide a streaming and non-streaming version of a pyrosetta.distributed framework unit test runner; all T900-series tests use streaming, except for 2 unit tests in T917 that use non-streaming since they spawn additional subprocesses that do not stream.
4. Catch and handle timeouts in non-streaming subprocesses launched in T917.
5. Refactor self-test.py away from forking processes, using customizable subprocess.Popen instead. Also closes an open file handle, pipes standard error to standard output for improved debugging, and prioritizes running T900-series tests first.
6. Run Python subprocesses in Benchmark server testing framework unbuffered.
7. Flush print statements in PyRosetta.
8. Micro-improvements to PyRosettaCluster such as better error messages, and temporary directory cleanup during exceptions.
9. Stabilizes the TestReproducibilityRemodelTaskUpdates test case by: (i) serializing double precision Python float values in Rosetta command-line options for precise round-trips through Rosetta C++ double precision core::Real values, and (ii) serializing PackedPose objects via PDB strings for updating task dictionaries between PyRosetta protocols.
10. Adds Python-3.8 compatibility in the TestReproducibilityTaskUpdates test case: random.randbytes(14) -> random.getrandbits(14 * 8).to_bytes(14, "little")

[lyskov]

Also, could you please update PR description to accurately reflect change set? (ie changes to self-test.py needs to be removed) Thanks,

[klimaj]

@lyskov I have updated the PR description with the current change set in PR #648, which supercedes this PR. Therefore, I will close this PR.