fix(kms): retry busy drm devices during startup

[FreddyFunk]

#2382 introduced a regression, which causes devices like my Framework 13 with an Ryzen 5 7640U to login into a black screen from cosmic-greeter.

On a greeter to session handoff the outgoing compositor can still hold the dri card when the incoming cosmic-comp starts, so the device open fails with EBUSY and the session dies with "Backend initialized without output" and a black screen. This retries failed devices for up to 5s at 100ms intervals in init_backend(), but only while no device came up at all, so a failing secondary GPU never delays startup.

• I have disclosed use of any AI generated code in my commit messages.
  • If you are using an LLM, and do not fully understand the changes it is making to the code base, do not create a PR.
  • In our experience, AI generated code often results in overly complex code that lacks enough context for a proper fix or feature inclusion. This results in considerably longer code reviews. Due to this, AI authored or partially authored PRs may be closed without comment.
• I understand these changes in full and will be able to respond to review comments.
• My change is accurately described in the commit message.
• My contribution is tested and working as described.
• I have read the Developer Certificate of Origin and certify my contribution under its conditions.

[Drakulix]

#2382 introduced a regression, which causes devices like my Framework 13 with an Ryzen 5 7640U to login into a black screen from cosmic-greeter.

Can you do regression testing which of the two commits from said PR caused this behavior for you? It was precisely meant to make this more unlikely by optimizing the shutdown procedure of the greeters cosmic-comp instance.

Re-trying is a really ugly workaround for something that should be considered a greeter-bug, so I'd rather revert parts of that PR, while looking into finding a better solution.

[FreddyFunk]

Yes, but probably in a week earliest. I'm now traveling without access to the hardware that has the affected AMD GPU. I'm 80% certain that e82a763 introduced the regression when I debugged this locally while working on this fix.