Skip CI when drafts are opened

[workingjubilee]

There is no particular reason to run CI against an unfinished PR,
since it is likely the author already knows it will fail. When
the PR is no longer a draft later commits will cause CI to rerun,
and bors exists, so this should be no problem for workflows.

Resolves #67357

[rust-highfive]

r? @pietroalbini

(rust_highfive has picked a reviewer for you, use r? to override)

[pietroalbini]

Nominated for discussion at the next infrastructure team meeting.

[Mark-Simulacrum]

I am personally against this change, I feel like it hurts rather than helps. I personally do not see skipping CI as useful -- if you didn't want feedback (both from humans and CI) then don't open the PR :)

That said, I could see an argument for disabling rust-log-analyzer from posting comments on draft PRs, because it does seem reasonable to presume that they will have more iteration.

But overall it feels like this is highly individualized: some people want CI feedback as soon as possible, others would prefer to wait. I'm not sure there's a one size fits all solution.

[jyn514]

Personally the reason I open draft PRs is because I've run out of time for the moment and I don't want to forget about them. So I'd be fine if CI didn't run: I know why it's a draft.

That said I agree this should be opt-in - is there a way to tell CI 'don't run this' in the PR itself?

[Dylan-DPC-zz]

I prefer the CI to run on draft PRs because it can pick up errors early in the lifecycle rather than show up when you are requesting a review which can delay a review further

[pietroalbini]

We discussed this in the infrastructure team meeting and decided not to go forward with this approach. You can see the context on the decision by reading the contents of the meeting.

Thanks @workingjubilee for proposing this!