abci: Make ABCI and P2P wire-level length delimiters consistent

[thanethomson]

Closes #9176.

---

PR checklist

• Tests written/updated, or no tests needed
• CHANGELOG_PENDING.md updated, or no changelog entry needed
• Updated relevant documentation (docs/) and code comments, or no
  documentation updates needed

[tychoish]

I think this makes the change more confusing: because what we're doing is just using varint encoding rather than encoding (effecitively) two fixed width integers. In this statement we say "here's the thing that we changed" and not "here's what you need to do (update your socket client implementation.)"

Reading this, my gut says "but aren't uint64 and int64 integers encoded the same on all twos-compliment machines?" and while this is true, the change is actually about varint encoding.

[thanethomson]

We were using varint encoding before, but we were varint-encoding an int64 value instead of a uint64 value. The two types of varint encodings produce different values on the wire, making them incompatible.

Client libraries (such as the one in Rust) usually support the uint64 varint encoding, and not int64, meaning that if anyone wants to interface with ABCI via TSP they need to roll their own length-prefixed decoding mechanism.

I'll clarify here.

[sergio-mena]

Could we mark somewhere that a similar change will be needed in the "abci++" folder that @jmalicevic is working on (see #9201)?
@jmalicevic , maybe you could extend the issue's description with an extra checkbox?