@joenepraat
Actually I don't understand 1., using multi device since ages with xmpp.

But maybe I understand it if you mean in the context of 2.?

And no, "xmpp" will never always use e2ee or even by default.
XMPP isn't a company but an extensible standard and client developers can make their clients freely how they like and for use cases that are not only secret private messages.
XMPP powers Apple and Google push servers and Nintendo switch status messages, so it isn't a thing to demand e2ee in the protocol.
Sayng that:

There are clients that default to e2ee.

I have the feeling many in the xmpp community don't have such a strong need for e2ee everywhere always, because many host their small servers themselfs ( it it easy and cheap) . And then you might have a different threat model, and don't like the inconvinience of PFS encryption so much.
Ultimately e2ee is for hiding your message from the server.

@joenepraat
Maybe think of xmpp like activitypub, rather then like, only mastodon.

It talks the protocol with the server but clients are infinite diverse with different features. They can generally talk on a basic level (OK, many more then on a basic level)
But they can be quite different in what they want to be.

For seamless regular chatting, I try to have the users on my sever (friends family) use a small bunch of all the exiting clients, that have basically the same feature set.
I can talk to anyone else, but then one must expect that some features won't work. (luckily the standard generally has fallbacks for everything, so one can notice that this)

@joenepraat
I think there was a module that enforces encryption on the server. And doesn't deliver unencrypted messages.. 🤔
In that case I would suggest you tell your users to only use very mature clients....
And you can still see metadata in any case. Best is not to spy on the users.
If that's impossible, xmpp Isn't the right protocol.

Generally xmpp uses many and rather small servers so one can choose an admin one can trust.

Because whatever services like for example signal say, it is always possible to get information out of metadata, by correlating time and IP etc. A sever always needs to know where to send what. And even if encrypted, timing can reveal a lot.