extend "Review Requested/Submitted" autolabeling to consider CI failures

[jyn514]

@edef1c mentioned that gerrit automatically sets PRs to waiting-on-author when CI fails and suggested we do the same. i think we should be a little careful about this - there are some cases where CI fails but the author doesn't know how to fix it (see S-waiting-on-help) - but broadly the idea makes sense to me and i think it would be nice if triagebot supported it.

[apiraino]

Slightly tangential question: after a perf run who should review the results? Currently we don't flip the review switch, example rust#140064. I personally think it's the PR authors' duty to investigate regressions but a perf-regressions-triaged should come from the reviewer.

[Kobzol]

That is likely what you want in the majority of cases, but there are also PRs that await a "vibe check" review (before CI is green) or that await a review so that PR can in fact become green. I'm not sure if we can be smarter than the reviewer/author setting the labels appropriately here.

Slightly tangential question: after a perf run who should review the results? Currently we don't flip the review switch, example rust#140064. I personally think it's the PR authors' duty to investigate regressions but a perf-regressions-triaged should come from the reviewer.

Perf. builds are requested explicitly, so I don't think that we need to modify waiting on author there; the originator of the perf. run comment likely wants to examine the results. Btw, when a try build fails, the PR will be flipped to waiting on author automatically.