[Kinflou]

Moves the simulator onto two runtime changes now on main:

• runtime#49: the jsonrpc feature builds JSON-RPC without std.
• runtime#50: encode_request_body on Framing and ConnectionFraming.

Changes

• comline-runtime is pinned to 207b38a, with features = ["alloc", "jsonrpc"].
• The JSON-RPC copy vendored in src/framing.rs is deleted. WireFraming::Jsonrpc now calls comline_runtime::framing::JsonRpcFraming.
  • runtime#45's found-* vectors in the tests still pass byte for byte.
• Preencoded and its test are deleted. That was the WireFormat that ignored its value so it could pass encoded bytes through.
  • WireFraming::encode_request now takes the body as &[u8] and calls each framing's encode_request_body.
  • It no longer returns a Result, because nothing in it can fail.
• serde_json no longer enables raw_value. Only the deleted copy used it, and the runtime turns it on for itself.
• src/framing.rs's module doc now says the simulator keeps its own request ids, because it plays both ends on one simulated clock.

one_way

Requests now pass the call's real one_way instead of a hard-coded false. This changes no bytes:

• datagram@1 and datagram@2 have no bit for one-way.
• The simulator's JSON-RPC behaves like a lockstep Client. It only models server-provided protocols, and the Rust generator puts such a package on a Client, not a Connection (on_connection in generation's connections.rs). Client::notify keeps the id.
• Only on a Connection is a one-way call sent as a notification with no id.

A new test pins this: one-way and two-way requests come out identical on every framing.

Wasm size (post-wasm-opt, from the CI logs)

The figures are in a comment below.

[Kinflou]

Wasm size after wasm-opt -Os, from the CI logs (main run 40226793179 vs this PR's run 40242337525):

build	before	after	change
scripted (pkg-script)	2,203,964	2,195,253	−8,711 bytes
lean (pkg, --no-default-features)	595,508	586,363	−9,145 bytes

CI is green.