The critical issue is there is a finite number of public IPv4 addresses. For example, mobile devices rarely get a public IPv4 address, most use CGNAT today. And even new services like Starlink, charge for a public IPv4 address since it's a finite resources. Some mobile providers are only IPv6 and relay on DNS64, which generally works since app store testing explicity checks for IPv6 DNS64/NAT64 compatibility.
The net result today is public IPv4 addresses have a cost, whether it's borne by provider or charged as add-on to consumer...point is they are not "free". And, there are other costs to end-to-end dual-stack stack network, since you're essentially managing two networks at all points within the infrastructure. So IPv4 translation at the edges of the network has potential to reduce costs.
Flip side, as a consumer of internet services, obviously getting both a public IPv4 and IPv6 prefix has a lot of benefit since you can access either network. But similar to providers... the end-user now still left managing two networks, and particularly needs to manage two sets of firewall rules, etc. In a simple, flat home LAN (e.g. no VLAN, one LAN segment), dual stack is pretty automatic. But once you start adding VLANs, managing dual-stack becomes non-trivial.
Anyway...the problem/request here is in a pure IPv6 networking RouterOS lacks any translation tools like NAT64/464XLAT/MAP-T/etc. this make using anything but a dual-stack architecture impossible.