The signal servers will absolutely have public and static IP addresses. You would not be able to connect to them reliably if they could change at any time and you had to rely upon DNS updates to find the server. AWS is not magic.

And yes, AWS has IP ranges allocated to it that they pull their public IPs from, that’s all that link is talking about — this page even provides the context that the IP ranges are available in order to identify which traffic is coming from AWS in order to allow / disallow it. Of course the AWS IP allocations won’t tell you which IP is associated with which service (and indeed many IPs, particularly in the IPv6 space, are probably not in use at all).

There’s not enough unique IP addresses to distinguish Signal servers

Why? Yes, IPv4 address exhaustion is a thing, and yes AWS only has a slice of IPv4 addresses to give, but you absolutely can get static public facing IPs from AWS that will be unique to your server. You can even pay for an elastic IP so you can hold a particular address and move it between instances. There is no way Signal does not do this.

Signal has native support for proxying via Tor in that case.

Yes, though the use case is mostly for getting around censorship. Realistically if you don’t want the government to know you’re using Signal… Do you want them to know you use Tor?