Single runs need to be able to recover from segfaults

[wesm]

This has been a problem for me-- some benchmarks have been mysteriously segfaulting that for past pandas revisions. Might be Cython-related. Would be nice of vbench would recover gracefully and skip the segfaulting benchmark