I didn't read all of the posts so this may be redundant.
Things may get better with AI. If they can make Elvis sing a Lady Gaga song it should be able to create whatever dynamics you want. They should be able to re-create the whole recording. (Overall, I'm not optimistic about the effects of AI on music.)
There are things in the real world that make "undoing" compression impossible (with today's technology). There is often compression on the individual tracks before mixing and there's simply no way to know the original peaks once the audio is limited. Other things make it practically impossible, even if theoretically possible.
If the compression parameters are known and controlled, it CAN be reversed. DBX noise reduction used linear 2:1 compression during recording and complementary expansion (including the timing) during playback for an incredible signal-to-noise ratio on cassettes. But they were unplayable without expansion. Dolby uses a bit of compression/expansion too, but not so much that it's unplayable if not decoded. The phone company also used analog "compansion" to improve the signal-to-noise ratio over the transmission lines. But these are different applications, and not necessary with digital.
The record companies (and most musicians) give the customers what they want. If more-dynamic music starts winning Grammy's, everybody will copy that style. But we've got more than one generation of listeners, musicians, and producers that have grown-up thinking that dynamics are bad and constant intensity/loudness is how music should sound.
I was around when CDs came out and I was expecting musicians & producers to start taking advantage of the improved dynamic range capability. I couldn't have been more wrong.
