Generate run ID per benchmark repo instead of build

[alistaire47]

This is a bit of work, but ideally we'd move from generating a run ID for each build to a separate one per benchmark repo (or internal to the Run in some fashion; per-call is really better than per repo in some contexts). This didn't matter before, but now benchmarks can start and finish a run, recording useful information there if something blows up, but existing runs can't be started and finished runs can't be continued, so they can't be shared across runners. For tracking we likely still want ABCI to know what all the run IDs are and pass them down instead of runners managing their own.