This gets suggested and rehashed on every single IPv6 comment section. There is no way to make IPv6 compatible with IPv4: by the Pigeonhole principle, there is simply no way to reference 2 ^ 128 objects using only 32-bits. The next usual hack is to "add an option" or something; but all the routing hardware would need to understand that option, which entails upgrading the entire Internet, which is exactly what IPv6 is doing/has done, except that adding an option ends up with a worse design.
Any suggestion needs to fulfill a. any machine should be reachable and b. you can't upgrade everything, or you may as well just go with what we've got.
In the end either all those things were backported to IPv4 or turned out to be bad ideas. IPv6 ended up being just more bits, but with a fully incompatible implementation.
If they had known this at the beginning, we could have transitioned to IPv6 long ago, by designing for a smooth transition period and reusing as much of the IPv4 stack as possible.
An end-to-end extension header would have been simple - something that can fallback to NAT connection tracking when it's not recognised. And it would have saved man-centuries of effort.
Coding for a mix of paths creates a 3 way test problem. You have to test purely legacy, the mix of legacy and new and then just the new. You suddenly have to start coding for runtime depended details.
The only thing that was the critical flaw in v4 was lack of address space. Yes, there are obviously a lot of other issues with v4, but instead of just fixing that one critical flaw, v6 suffered from classic scope creep, put in the kitchen sink, in a way incompatible with v4, and 2 decades later we're still 5 years away.