These are interesting problems, and we've certainly had a few debates at the Telehash meetups about the meaning of "identity" and such. Hashnames aren't expected to be human-friendly. Ultimately, I think it's a problem to be solved at a higher layer than Telehash.
Telehash is an online protocol, in much the same way that IP is an online protocol, so message store-and-forward isn't in scope. However, such a system could be built on top of Telehash. In fact, developing the design for just such a system has been on my mind quite a bit lately. ;)
From what I understand, hashnames aren't supposed to be like domain names. They're supposed to be like IP addresses. Normal people don't need to type or remember IP addresses very often.
So what it needs is a DNS replacement. Something like namecoin might work.
I think friend discovery is an abstraction layered on top of the protocol itself. Consider the address books on every personal device. In fact, most of the popular communications protocols/infrastructure of the last century has required its own social abstraction.
One problem I saw was that the hashnames are still too long and random to be practical for many use cases like building instant messaging on top of it. What you kind of need on top of it is some kind of possibility to find other people/hosts. One possibility is kind of a social network way: You show your "friends" who your "friends" are. Hashnames that are friends to many of your friends are most likely friends and could be suggested to you. Another possibility is local node detection. Other devices you own are most likely on the same network.
Another problem is that sometimes instances aren't online but still should be able to receive messages. You either need to persuade some nodes somehow to store messages or you need central servers.