You seem to fundamentally misunderstand what "end-to-end-encryption" means
"End-to-End Encryption" is an approach different to "Client-Server Encryption", and is relevant when users want to communicate with each other, without the server knowing what they are communicating about.
Client-Server Encryption Explained
Alice uses the amazing service "example.com" to send a message to Bob. Alice encrypts the message on her end and sends it to example.com. The server then decrypts the message, sees that it belongs to Bob, and encrypts the message again, sending it to Bob.
This means that the server at "example.com" can see the content of the message. This can be a problem for Alice and Bob, if, for example, they live in a country where the government is known to spy on its citizens. It is possible that the government forces the owners of "example.com" to screen all messages for specific keywords and report all messages, including sender, recipients and contents to the government.
While a Client-Server Encryption model still protects against attacks "on-the-wire", it doesn't protect against attacks that compromise the server.
End-to-End Encryption Explained
Alice uses the new and improved amazing service "foobar.com" to send a message to Bob. "foobar.com" uses end-to-end encryption, instead of a client-server model.
Alice begins by signing her message with her private key, then encrypts the message with Bob's public key. Alice then encrypts that message again via TLS and sends it to the server. All the server can see is that it is a message from Alice for Bob, but not the contents. The server then encrypts the message again via TLS and sends it to Bob. Bob is now able to decrypt the message with his private key and verify that it is really from Alice, by verifying her signature with her public key.
This scheme can also be modified slightly, by removing Alice's signature and thus making it more difficult for an adversary intercepting the messages (and possibly forcing Bob to decrypt his messages) to see that Alice is the original author.
The advantage in this model is that it becomes much more difficult for a nation-state actor to do widespread surveillance. While it is not impossible for them to attack high-value targets, simple keyword searches as described above are not feasible anymore.