Earlier quoted context omitted.
You can implement search on top of crypto hashes. This is how decentralized P2P file networks like Kad worked. The discovery doesn't need to be at protocol level. You can have some servers following large number of accounts, analyze the feed, create a "trending" list and publish that under their own account that you follow. Then you can have second level "trending" servers which aggregate over first level ones if you…
It makes sense to me to take content discovery out of the protocol. This allows different apps on the same network to create different mechanisms for discovery. Some may want a pure chronological view of posts from people the follow with no algorithm, some may want pure algos a la TikTok. It seems like the nostr protocol allows for all of this. Users can decide their experience by downloading different client apps.
There's a big difference between creating a protocol that's intended for fully-content-agnostic communication over the Internet (eg, HTTP) and creating a protocol that's specifically intended to be used to build a social network. The latter type needs various features that enable the type of behavior we know people on social networks desire from them. It does not, as you say, necessarily need to encode strong opinions about how content discovery should be done into the protocol at a fundamental level, but it at least needs to acknowledge that that is a thing that will be necessary for any social network to succeed.