[stress] Diagnostic output from script should be refined

We recently ran into this with @HarshaNalluru.

There was a compilation failure in his stress test, which was easily diagnosable but there was so much diagnostic output that it wasn't easy to see where the failure was. Once we found the error line from the docker build (when it compiled his code) it was obvious, but getting there wasn't as easy as we'd like it to be.

We should be able to improve this a bit, possibly by just re-echoing the most relevant command failure's output at the end, rather than just printing out the failed command line, which is what we do today.

The issue names no files or tests; start by locating the script that runs stress tests and captures docker build output. Reproduce a compilation failure and inspect how the failed command's output is currently displayed. Done means the relevant failure is easy to find at the end of the diagnostic output.

Written by the indexing model from the issue text.