It's never worked like that, you just provide an arbitrary address, Collisions are not really an issue because it's a manual thing done by a professional for one of any number of uncommon but legitimate reasons.
Assuming its not a bug, to me this seems like yet another small but meaningful step in turning macOS from a general purpose OS to a consumer only OS with all the "most people don't want or need this" stuff taken out.
I don't think they are. They are saying that if you are connecting to an Ethernet network which is only being used for a /24 IPv4 Network, then that Ethernet has plenty of MAC addresses available.
MACs don't need to be unique outside of a given bridge domain and in the modern world there is almost always 1 bridge domain per subnet. It's actually fairly common to have "duplicate" MACs in enterprise networking, well known MACs are reused for first hop redundancy protocols like VRRP.
MACs are 48 bits, and the unique ID is 24 bits if you preserve your real OUI. This doesnt have anything to do with with an IPv4 address.
Furthermore, "professionals" rarely randomize their IPv4 address; theres almost no value in it. The subset of people with some kind DHCP filtering but no NAC or DHCP enforcement is miniscule.
If a subnet has 254 machines, they are all made by the same vendor, and the last 3 bytes of the mac address are randomized for every one...
There's (254*253)/2 combinations of machines.
Each one has a probability of about 1/16777216 of a collision with true random numbers.
The number of collisions you expect in this scenario is ~0.0038 over the network. With the wacky problems that happen on MAC collision, this is rare but still too common for routine use.
But the chance of any issues, if now and then a few professionals randomize their address, is basically nil.
Assuming its not a bug, to me this seems like yet another small but meaningful step in turning macOS from a general purpose OS to a consumer only OS with all the "most people don't want or need this" stuff taken out.