service: spawn body reader via wasm-bindgen-futures on wasm32

[connyay]

spawn_body_reader runs the per-request envelope decoder in the background so bidi and client-streaming RPCs can read incoming messages while the handler is producing responses. It used tokio::spawn, which panics with "there is no reactor running, must be called from the context of a Tokio 1.x runtime" on wasm32-unknown-unknown targets (Cloudflare Workers, etc.), where the ambient executor is wasm-bindgen-futures rather than tokio. Server-streaming and unary RPCs worked on Workers, but bidi and client-streaming panicked the moment the first envelope was decoded.

Under cfg(target_arch = "wasm32"), dispatch the reader future via wasm_bindgen_futures::spawn_local instead. spawn_local doesn't return a joinable handle, but the existing code only held the JoinHandle for lifetime association (don't abort early), so None works fine. spawn_body_reader's return type and
StreamingResponseBody::with_reader_task now take Option<JoinHandle<()>>. tokio::sync::mpsc keeps working unchanged; its internals only need wakers, which wasm-bindgen-futures provides.

Adds a wasm32-only target dep on wasm-bindgen-futures = "0.4". No behaviour change on native targets; the tokio path is unchanged.

Verified end-to-end against a Cloudflare Worker via miniflare and wrangler dev: bidi calls now succeed, the handler is dispatched before the request body completes, and per-input echoes round-trip in ~1ms while the request stream is still open. See
https://github.com/connyay/example-connectrpc-worker for a runnable example (clock = server-stream, echo = bidi, heartbeat = full-duplex bidi probe).

Scope here is server-side handler dispatch. call_bidi_stream in client/mod.rs also calls tokio::spawn, but the rest of the client module already requires hyper / tokio-net features that don't compile on wasm32. Wasm client transports are a separate concern handled by user-supplied ClientTransport impls (see examples/wasm-client).

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[connyay]

I have read the CLA Document and I hereby sign the CLA

[iainmcgin]

[claude code] Pushed a small fixup commit (d0ce6b0) extracting a spawn_detached helper to centralise the platform fork. The two inline #[cfg(target_arch = "wasm32")] blocks in spawn_body_reader are replaced with a single let reader_task = spawn_detached(reader_future); call, with the cfg gates moved onto two definitions of the helper (one with the Send + 'static bound for tokio::spawn, one with just 'static for spawn_local). The wasm hedge in with_reader_task's docstring also drops out — the platform contract now lives in one place.

Verified locally:

• cargo check -p connectrpc (native) — clean
• cargo check -p connectrpc --target wasm32-unknown-unknown --no-default-features — clean
• cargo test -p connectrpc --lib — 235/235 pass
• cargo clippy -p connectrpc --all-targets -- -D warnings — clean
• cargo doc --document-private-items — same warning count as baseline (9 pre-existing intra-doc-link warnings unrelated to this PR)

Happy to revert if you'd rather keep the original shape, but this gets the diff to a place we're comfortable approving and merging.

[connyay]

The fixup looks great, thanks!