Better observability tooling for big runs

The problem:

Currently, the output of Modelbench during a run is primarily progress bars showing the progress of test items against a SUT. This is moderately useful when running everything sequentially, but especially as we attempt to run these jobs in parallel, this becomes difficult or impossible to parse.

The solution(?):

Well, we're not sure yet. But, ideally the solution would let us do things including, but not limited to:

• being able to see the progress of test items against all SUTs, regardless of whether they're being run in sequence or in parallel and;
• being able to see the overall progress of the benchmark run and;
• being able to see how long the run has been running and;
• being able to see, maybe, very approximately, how long the run anticipates continuing to run and;
• being able to see where the run has encountered problems, what kinds of problems those are, and maybe if there's a way to resolve them

I've done some very rough sketching with a library called Rich to see if it'd be useful and came up with something like this (which was mostly working-ish), but I don't think anyone is married to this particular idea:

Start by reviewing Modelbench's current progress-bar output during sequential and parallel runs, then inspect the existing Rich sketch referenced in the issue. Done means an agreed observability approach that shows benchmark and test-item progress, elapsed and approximate remaining time, and run problems with useful resolution context.