Instead of having one struct with uninitialised fields
which are mutated, have explicit states and ensure that the types
show that there is no ambiguity which data is available or not.
Consequently, this removes quite a few `unwrap`/`expect` calls.
* Use the sha2 crate in the handshake
* Return a Digest in algo_support instead of a ring ref
* Switch to ed25519-dalek for keys
* Make ring more or less optional
* Switch to ed25519_dalek for the verification
* Extract the key exchange to its own module
* Remove the ring RNG from the handshake
* Some warning fixes and forgot file
* Move key exchange to own module
* Remove usage of ring::digest
* Remove ring from handshake entirely
* Implement ECDH for WebCrypto
* Remove the libp2p-secio feature
* Fix ring being included
* Address some concerns
* Provde some panics in WebCrypto
* Prove the Hmac panic
* Prove more panics
* secio: keep ciphertext if start_send is not ready
Otherwise ciphertext may be thrown away and once sent, the cipher states
have diverged, resulting in nonsense messages on decryption side.
* Also attempt to send pending data `Sink::close`.
* Inline `send_pending`.
* Introduce NULL cipher and allow more configuration.
* Back to using the hash-code for handshake.
Using `Endpoint` would be incompatible with the existing protocol.
* Add comments.
