substream/fix: Allow empty payloads with 0-length frame

[lexnv]

This PR ensures that zero-length frames are propagated to higher levels.

An issue was noted with the compatibility between litep2p and smoldot, which resulted in smoldot not being accepted by substrate-based chains.

The issue stems from the litep2p substream (high level) implementation, which did not handle zero-length frames.

Smoldot sends during the notification handshake a zero-length frame (ie, frame [0]).
However, full nodes (litep2p and libp2p) send the node role Role::Full (ie, frame [1, 1]).
Substrate nodes expected the node Role, however zero-length frames were never punished.

This behavior causes the zero-length frames to go unnoticed and further call into poll_read.
In the case of smoldot, the connection is terminated after the handshake timeout of 10 seconds.

Testing Done

• Reproduced the issue locally with zombinet and papi (https://github.com/polkadot-api/polkadot-api/tree/main/integration-tests/zombie-tests)
• The e2e tests were timing out, because litep2p rejected the /transactions/1 substream with smoldot (however it accepted the /sync substream which sends a different payload)
• Added a new notification test to illustrate this behavior with different payloads (including empty ones)

Closes: smol-dot/smoldot#2128

Thanks @josepot and @tomaka for repro-case and details 🙏

[tomaka]

Smoldot sends during the notification handshake a zero-length frame (ie, frame [0]).
However, full nodes (litep2p and libp2p) send the node role Role::Full (ie, frame [1, 1]).

Ah, I see.
If I recall correctly, my intention as the designer of the protocol was that notifications have an empty handshake. In Substrate, however, at the time the code written there was no way to differentiate between transactions and GrandPa in the code that was sending the handshake, so I simply sent a Role all the time. I guess I overlooked this when writing smoldot.

I'm not really opinionated on this, for what it's worth, and I can also modify smoldot to send a Role if desired.

[sistemd]

LGTM!

[lexnv]

Sorry for the late response here🙏

Indeed this was a miss on the litep2p side. Assuming it's a minor change, sending the role of the light-client (https://github.com/paritytech/polkadot-sdk/blob/6ac204d52beab2e5c22588519f639cc03781ec26/substrate/client/network/common/src/role.rs#L93) would still benefit Kusama, which uses litep2p entirely. We'll follow up with a release shortly, but I expect a window of 2-3 weeks until enough nodes update to the fix.