I am also very unhappy with the direction Signal has gone, but there's currently no alternative. I'd be interested in contributing to work attempting to replicate it, though.
It's such a relief to have a modern chat system that's FOSS, federates, and offers a liiiiitle more advanced UX than irc...
Encrypted attachments was PRed this week: https://github.com/matrix-org/matrix-react-sdk/pull/533 (for the web; mobile will follow shortly)
Disclaimer: I do a lot of work on XMPP related things and rather like the protocol, so maybe I'm biased. I'm sure there are things for which Matrix is a better fit that I wouldn't use XMPP for too though.
Matrix is not "all pull-based" - that's just the baseline implementation. Folks have already implemented push-based transports for Matrix - e.g. COAP or WS.
edit: [Disclaimer: i work on Matrix]
Other projects can adopt parts of the Signal Protocol, and doing that is certainly better than just making stuff up. But Signal owns the protocol, because the experts who own the protocol are attached to Signal.
It's not connected to Signal or Signal Protocol or Open Whisper Systems, and subsequently we've added an entirely different new ratchet (Megolm - https://matrix.org/docs/spec/megolm.html) to handle Matrix's specific requirements for E2E.
But the reason crypto people are so positive about Signal Protocol isn't just that they like the ratchet. It's also that they trust the entire design of the system, not just the ratchet but all the rest of the cryptographic details, and also the oversight of the protocol.
It's kind of the same way that crypto engineers like stream ciphers designed as simple hash cores running in counter mode, but really what they like is stuff that Dan Bernstein designs --- they aren't encouraging you to go design your own hash-core based stream cipher!
So: it's good that these other systems adopting "Axolotl" are at least starting from a cryptographically serious place. But it's a bit jarring to see them reference "Axolotl" as if it answered the question of "why should we trust this cryptography".
A better answer would be to provide the bios of the people who designed and implemented the crypto in these systems.
In terms of bios: the folks working on libolm have 10-15 years each of professionally writing decent security-conscious native code, the vast majority of which (pre-Matrix) has been proprietary, with the exception of occasional contributions to things like Wireshark. I don't think they'd have described themselves as specialising in cryptography before embarking on libolm, but the team's learnt a lot along the way and the label might be more appropriate now. Ooi, what would you consider an appropriate bio? (short of being DJB or Moxie? :)