Debugger Remix: Terminal outputting ALSA function 'snd_pcm_poll_descriptors' failed with error 'UnknownErrno: Unknown errno'

## Bevy version

0.13

## \[Optional\] Relevant system information

* Arch linux
* Stable Rust 1.77

## What you did

(Sequel to #9798)

So for the past few versions, I've been experiencing the error described in #9798: Every now and then there will be rapid spamming of stdout with messages about audio fails until I restart the game. I'm developing a TUI game, so its rapid outputting to console was much more problematic than it would be for a GUI game. It also means that while I plan to have audio included in the game, it is an expected use case that users might want audio disabled. So while the workaround proposed in the original issue is to disable audio features, I actually do use the audio features, but I don't have background music yet so I've only been using it for SFX, hence why it triggers so often. I've recently tried other workarounds, such as creating an audio that plays a silent mp3 on loop, or playing an actual audio file in loop on low volume, but neither solution worked.

Then recently, I also recently started setting up a debugger with my IDE (codelldb via neovim+dap). Whenever I attach my debugger to my program, even if I don't hit any of my breakpoints, it'll start spamming this message every time. I even tried it WITH audible background music and the issue still occurred. This makes me feel that even the fix proposed for the linked issue of deferring audio stream initialization wouldn't prevent this problem from happening.

## What went wrong

I am hoping to find either a better workaround for audio where I can prevent this from happening both in normal use and in the debugger that doesn't require me removing all audio features for my game

## Additional information

Version of codelldb I'm using: codelldb 1.10.0-1 from AUR

Otherwise, very similar to #9798, which was closed as "won't fix," since there seemed to be an acceptable workaround for that user and the fix introduced unacceptable latency regressions. I feel since neither the workaround nor the proposed fix would likely handle my use cases, I thought it was worth re-opening (Let me know if I should've done something different).