The part about PFS falls down, for me, when you assume both that keys can be cracked

The argument doesn't hold because its known that keys aren't 100% secure [0]. This is why recommended key algo's and lengths are constantly changing and increasing in size because either they're found to be weak and/or compute has increased to the point where, although it may not be easy, it's still a possibility for someone with enough resources to break. And again, at least on the Signal side, they're left with having to decrypt every single message, where with Session, one key gets you access to everything.

I do agree that disappearing messages should be enabled for maximum security.

[0] See page 54 in this PDF -> nvlpubs.nist.gov for what NIST recommends for key length and specifically note the caveat that:

The security-strength estimates will be significantly affected when quantum computing becomes a practical consideration.

Page 62 further presses the point:

At some time, the security strength provided by an algorithm or key may be reduced or lost completely. For example, the algorithm or key length used may no longer offer adequate security because of improvements in computational capability or cryptanalysis.