Add failures to the end of the output #1201

When running asv run, if anything fails, it's only available in the middle of a possibly huge log. Finding the errors can be quite time consuming, and it'd probably be helpful to have them repeated at the end of the log, similar to what pytest does.

@mroeschke implementing this in asv would be equivalent to what you proposed in pandas, right?

xref pandas-dev/pandas#46599 (review)
xref https://github.com/scikit-image/scikit-image/blob/main/.github/workflows/benchmarks.yml#L81