This is a great question and it brings up some subtle points of how EIGRP functions. Queries are subject to the split horizon rule just like advertisements are. If you get a query for a network from an interface, you never send out a query for that same network from that interface.
So in the first case, when R2 lost connectivity to the loopback network, it sent a query to R1 for an alternate route to that destination. This is because in this case, R1 is not a stub. Because R1 got a query from R2, it did not send a query back to R2 via that interface due to the split horizon rule.
Now in the second case R2 knows that R1 is a stub so it doesn’t sent a query for the lost network. When R1 learns of the lost connectivity to the network (via updates etc), it hasn’t received a query for that network, so it can send a query to R2 without breaking the split horizon rule.
I hope this has been helpful!