You need to check the experience of distributed networks like I2P or GNUnet. Maybe something like the GNU Name System? Any decentralization of DNS leads to the use of DHT. But any implementation of DHT requires an initial download of data about the network nodes. This is always the bottleneck in such networks.
Nostr IS already a DHT.
The "H" in DHT stands for "hash". Pub keys aren't hashes.
event ids are
This is not really a bottleneck. You initially download only a small number of node informations for bootstrapping.
The bottleneck is not in terms of the volume of data being loaded, but in terms of "centralization". Node data need to be kept up to date all the time. There is also a vulnerability in the substitution of loading nodes. Not that these are insurmountable problems, they just need to be kept in mind.