When you do a IPv4 "pinhole" port map you get precisely one mapping and it will timeout eventually, which can lead to all sorts of exciting debugging opportunities.
IPv4 does have 1:1 NAT but there are so few IPs so whilst I have a /24, 2 x /28 and 2 x /29 to play with and others, I doubt most do.
NPTv6 maps an entire address space from A->B and is actually not designed to deal with NATv4 anyway. It is for outbound connections.
This is implementation specific: on my Asus I can specify to allow in all ports from in via a NPTv6-ULA hole, a port range, or even a single port.
If I want to only allow tcp/25 in to a particular IP I can do that, if I want 5900-5910 I can do that too, as well as 1-65535.
A router moves packets from A->B and a firewall defines what is allowed from A->B.
ULA to NPT is router stuff: NPT literally means "Network Prefix Translation" it turns all your IPv6 addresses into a normalised one and shifts them.
We are not talking about ports or protocols yet, just (IP) addresses.
Why? This is exactly how I want it to work: pass through only the ports (or all the ports) I want/need for the service in question.
One convenient thing about ULA+NPTv6 is that, unlike IPv4 NAT, is that an external IPv6 address is basically 1:1 'equivalent' to an internal IPv6 address. The NPTv6 is stateless so that, firewall rules allowing, a connection can come right in without all sorts of contortions for port mapping.
Most (residential) gateways by default block incoming requests unless they're a reply to a previous outgoing connection. I know I can ping6 the iMac I'm typing this on, but attempts to (e.g.) SSH in are blocked by default.