Issues already marked stale are clogging the stale action

[cameel]

After the update in #13947, the stale action started nicely by tagging batches of ~20 issues as stale per day. However, now it's gradually tagging fewer and fewer issues. There was only one marked stale yesterday. We need to fix that.

Looking at the last run, here's a breakdown of the 100 issues it processed:

• ~25 issues were exempt due to labels or milestones
• 1 was newly marked as stale
• ~5 were found not to be stale
• The rest was already marked as stale but not yet old enough to close

It's the last category that seems problematic. For these issues it prints 2 operations consumed for this issue in the log. I think it's stopping when it uses up its allotted number of Github API operations and that happens before it gets to non-stale issues. There's no such message for issues exempt due to labels or milestones.

This seems to be a known issue with this action: actions/stale#792. Looks like it can't cope with too many stale issues at once. I think that in the long term it would eventually get to the newer ones, once some of the stale ones get closed off, but having it stall like this is annoying.

Fortunately our stale period is not that long, so after 7 days it should start closing old issues and then it will be tagging as stale again. In fact, it should be starting to close the issues it marked stale first about now. Still, we should try to find a workaround:

• At the very least we should bump the number of API operations it's allowed to perform. It's the quickest and simplest thing to do so it's worth trying, but at some point we'll be rate limited. If we don't, it's great because the problem should resolve itself when we're done with our huge stale backlog.
• We could reconfigure the action to go over stale issues first and only then try closing them. Is that possible?
• We could do another period of more aggressive configuration, as we originally planned to, to get through the oldest issues faster.
• Any other ideas?

[cameel]

We discussed this on one of the recent calls and decided to limit this to doubling the operation rate (#14051). The bot is processing fewer issues in a batch than we'd like but it also does not actually get clogged so it should be enough.

I just merged #14051. We can close this issue if we're satisfied with the effect.