Don't put nullptr into bus details pointer, fixes segfault on volume change

[ellenhp]

I think this fixes an issue brought up by @AndreaCatania in the comments of #51296. It was introduced by me in that PR, and this is my first shot at fixing it.

It was simpler for me to think about the process of updating the bus and volume information of a stream playback when the audio thread took complete ownership of the bus details structure that was active, but I couldn't figure out how to fix this bug in a reasonable way without rearchitecting that a bit. So now instead of inserting nullptr into the bus details pointer to prevent the main thread from freeing the bus details struct, the audio thread makes a local copy of the bus details. This does mean that the audio thread is technically vulnerable to a use after free if the system scheduler decides to sleep it between lines 385 and 387 for long enough that the the bus details struct gets updated and the old version gets free'd on the main thread during SafeList cleanup.

Speaking as someone generally who's very into safety at all costs, I really don't think it's worth planning for that happening.

[AndreaCatania]

Thanks for the PR, I can confirm that this fix the issue. As momentary fix is great, however it would be great if you could improve it afterwards. Anyway, thanks a lot!!

[ellenhp]

Cool. Basically the idea here is that we're preserving the behavior where the bus details structure is never modified once it's applied to a playback, only replaced. It's a bit less obvious that it has that behavior now, but I'm not sure there are many obvious ways to improve it. I think that the potential use-after-free if the system sleeps the audio thread at the exact right moment for nearly an entire frame is an absolute non-issue. There are so many places where things like that could cause problems but it's just so unlikely.

[ellenhp]

I'm actually going to make this a draft because I want to investigate whether I could just replace all of my bus details pointers with values. I think when I built this to use pointers it was because the bus details struct was not trivially copiable, but I think it is now. No point in having all these pointers when I could probably just have an atomic that holds the entire structure.

[ellenhp]

I take it back, the bus details struct is not trivially copiable. Marking as ready for review.

[mhilbrunner]

cc @reduz

[ellenhp]

This may be superseded by #52237. I made the same changes there.

I can split them up but it made sense to fix all outstanding bugs in the polyphony PR to make testing easy.

[ellenhp]

Superseded by #52237