Thanks for doing this. I guess these constants should have never live the runtime environment and be serialized anywhere. IMHO, if there's a need to serialize a constant, then this is a hint that you should not use Constant Sorrow for it.

One more consideration before I finish this.

Our encryption looks like this: pre.encrypt(recipient_pubkey_enc, sig_header + plaintext)

We are adding this "signature header" to the actual plaintext.
The signature header was 8 bytes, now we are going to go with single bytes.

But why are we adding this stuff to the plaintext at all? Isn't it kind of a hack?
Would it be better for payload signature to be handled at the app layer?
Would you even trust NuCypher's code to sign your data if your data was so important?

I as an implementer might prefer my own signing mechanics on the the payload anyway and hand the data over to Enrico already signed.

Any support for removing signatures entirely?

For what it's worth, I agree with removing the signatures. I always thought of Nucypher as KMS, while ciphertexts are distributed via side channels, so, as you said, it's up to the application to decide whether to sign them or not. But I may be missing some attack vectors here - @tuxxy , @cygnusv ?

I'd love to see this PR finished, it's a nice prerequisite to our future plans with entity serialization.

Closing this PR for due to staleness. Leaving #1484 open.