It's an interesting project, but I think a clarification would be important: Is SSH3 supposed to be "SSH-over-HTTP" that happens to use QUIC as a transport, or is it "SSH-over-QUIC" that happens to use HTTP as an auth/adressing layer?
The difference is not just philosophical, it also has practical implications, in particular what part of the different protocols (SSH, HTTP, TLS and QUIC) clients, servers and intermediates are expected to implement and which can be left out.
E.g., if it's "SSH-over-HTTP", I'd expect the protocol to work well with HTTP proxies and application servers and I'd expect to be able to run a SSH3 server and a regular HTTP server on the same port. On the other hand, I'd expect features that require precise control over the low-level QUIC connection - like UDP port forwarding and session resumption - to be less reliable.
If it's "SSH-over-QUIC", the expectations would be the opposite: That you can treat QUIC (and TLS) as an integral part of the protocol, in the same way that the encryption, auth and transport layers in standard SSH are seen as an integral part of SSH. However then the server should generally be deployed as a standalone process on a separate port and should not be considered a fully compatible HTTP endpoint. That might diminish the "stealth" ability of the protocol a bit.
Or to sum it up, which parts of the protocol stack would an SSH3 client or server be expected to provide by themselves and which parts would be delegated to the OS/infrastructure/intermediaries etc?