Just because it’s centralized doesn’t mean that it falls under this risk sector. Theoretically if the app was open sourced and was confirmed to not share your private key remotely on generation (or cross sign the key to allow a master key…), then the most the centralized server could know is your public key, the server wouldn’t have the ability to obtain the private key (which is what is needed to read the e2e encrypted messages)

This process would be repeated for the other party. The cool part of that system is you can still share your public keys via the centralized server, so you wouldn’t need to share the key externally. You just need to be able to confirm that the app itself doesn’t contain code to send your private key to the centralized server. Then checking integrity is as easy as messaging your friend to post what their public key is, and that public key would need to match the public key that the server is supplying as your contact.

The server can’t MiTM attack it because the server has no way of deciphering the message in the first place, so the most it could do is pass the message onto the proper party whom has the private key to be able to decrypt it.