It's not intuitive, but it's also not easy.
I'm honestly not sure how to handle this. The stations will use whatever DNS they're configured to use, which by default is Untangle. Now Untangle uses all the DNS addresses listed on any marked WAN interface, in a round robin. This behavior is identical to most resolvers.
Now DNSMasq (DNS and DHCP service on Untangle), will resolve the address using whatever it has, and then pass it back to the client.
Now, with TunnelVPN you're asking Untangle to resolve DNS queries it's receiving, from specific devices, using specific servers. DNSMasq just flat doesn't have that ability, heck I don't know of a single DNS server that does.
So you then have to configure the DNS resolvers on the clients to use the specific DNS addresses via DHCP, OR use a forward rule to push the traffic to a specific location. This solves the "leak", but it also disables Untangle's own DNS functionality that many networks, my home included, depend on. And if the tunnel goes down, all those systems either configured to use the tunnel DNS via DHCP, or forced there via the port forward won't work because no DNS resolution.
Untangle is all about things working, everything above is full of stuff not working. The reality here is this use case demands different DNS resolution stacks based on which portion of your network is resolving, and when. And honestly, there is no clean way to automate all of that because the network admin has to determine all that.
And for what? There is no "security" added by using a VPN provider. The only thing you functionally get is the ability to access geo locked content. The VPN providers can claim they don't respond to government queries, but they are required to by law. And if they aren't required by law that means you're dealing with a nation that doesn't have the legal foundation to be trusted to begin with. The rest of our security is already handled by basically everything being SSL'd these days.
All of this is bonkers to me, but that's also why the product works as it does. And as you can tell I've spent considerable time thinking about this, and yet I don't have any good solutions. The port forward is the cleanest, and perhaps that could be an option added to the capture rules. But to really be a solution it needs to be able to gracefully fail back to using the ISP DNS or whatever is on external when the tunnel is down.