CI: Simplify call to asv

[datapythonista]

The way we are calling asv in the CI is too complex for no reason. Removing the next unnecessary things:

• asv check not worth calling, since just imports the benchmarks which will be later imported anyway when calling asv run
• Adding the remote to git doesn't have any effect I think, since we don't specify the commit, and we should be running benchmarks for what's been checked out
• asv machine is not useful, since we won't save the results, or compare them with other runs
• asv dev just avoids creating environments, but we can speed up things a bit by just running each benchmark once, and not save the results, which are not used anyway
• The conditional to just run the benchmarks if the previous step build is successful, is the default. It makes sense in other parts of this file, since there are several steps depending on a build, but not in this case
• I don't think publishing the asv log makes a lot of sense. I think I added that myself in the past, but now feels just a waste of time and resources (unless someone used that artifact, then please let me know)

I'm going to see how to remove the hacky grep and just leave the asv call, but I think this needs a new asv release.

[mroeschke]

LGTM.

Related, It would also be nice for the failing lines in benchmark.log to be echo/cat-ed to make it easier to pinpoint which benchmarks failed.

[jreback]

thanks @datapythonista