I thought I also saw BMP in a lot of games for some reasons (not to mention it's default output for the default image program on the effectively default consumer PC OS :)
[0]: https://gdal.org/drivers/raster/cog.html#general-creation-op...
Even 3D games will sometimes use them in a lot of odd places because it was faster/easier/took fewer cycles to include an uncompressed or simple run-length encoded bitmap than to try to store a smarter image format and need time to decompress it. (Though yes modern GPUs often support image decompression directly for certain texture file formats and that's shifted a lot.)
(Some of that I think is also political/contractual in the way that using MP3 would give some additional types of PC players "free soundtracks" while using other compression formats is a bit of security through obscurity and raises the bar slightly for PC players that don't, for instance, know what Ogg Vorbis even is.)
Do they?
Even if they do, it's typically useless for an application like a game, because the game will need to filter and/or mix that audio in ways that the hardware probably can't handle on its own. (Sure, maybe your hardware can decode one MP3 at a time, but can it decode two dozen MP3s at once, apply a bunch of environmental filters to them, and mix them down to a single stereo output? Probably not.)
The games have to deal with multiple layers of audios at the same time. So imagine one sound file have the dialogue for that specific scene, other sound file have the ambient music to compliment the surrounding, other file have sound effects, etc. So the games have to handle maybe 10 audio files at the same time which can put the strain on the CPU since it need to handle the decoding for compressed format. When that occurs, it eats up the CPU resources when it should be allocated for other stuff. Also don't forget some game have 5.1, 7.1, Dolby variants, etc which can be taxing on the CPU if it have to process multiple compressed audio at the same time.
Do they? Most of "modern audio" is just software decoding/mixing to a pretty not-smart-Digital-to-Analog Converter.
My educated guess is that most compressed audio formats are optimized for size, not for low latency recording/playback.
Also one of the reasons why Opus is so good and is replacing lots of older formats.
When making a quick hack like a toy ray tracer or Mandelbrot generator in C or C++, it's traditional to use .ppm [0] because it's trivial to write out pixel values in ASCII.
That works great for compressed textures like DDS or ASTC.
Unfortunately, the hardware decoders for proper codecs like jpeg or h.264 typically come with APIs prohibitively complicated to use, unless the app is a video player and needs to consume these APIs anyway, despite the complexity.