Switch wasm test/bench runner to wasm-harness

Summary

Replaces the wasmtime --wasi threads runner with wasm-harness, which executes wasm32-wasip1[-threads] binaries under a chosen engine (wasmtime, V8/d8, SpiderMonkey/sm) via a bundled WASI polyfill. Enables running cargo test / cargo bench against the same JS engines used in browser-targeted benches, without source changes.

Changes

• .cargo/config.toml
  • Added a [target.wasm32-wasip1] entry mirroring the threaded target's rustflags (atomics, bulk-memory, mutable-globals, simd128, 4 GiB max memory).
  • Both wasm32-wasip1 and wasm32-wasip1-threads now use runner = "scripts/wasm-runner.sh".
• scripts/wasm-runner.sh (new, executable)
  • Auto-detects host core count (nproc on Linux, sysctl hw.ncpu on macOS, fallback 1) and exports WASM_HARNESS_ENV_RAYON_NUM_THREADS=<cores>.
  • wasm-harness forwards env vars prefixed with WASM_HARNESS_ENV_ into the guest with the prefix stripped, so rayon sees RAYON_NUM_THREADS without polluting the host shell.
  • Pre-existing WASM_HARNESS_ENV_RAYON_NUM_THREADS is respected (user override).
  • execs wasm-harness with cargo-supplied args.

Usage

Prerequisites: cargo install wasm-harness. For JS engines, npm i -g jsvu && jsvu to install v8 / sm shells. With no engine selected, wasm-harness picks the first available from $PATH / ~/.jsvu/bin.

cargo test  --target wasm32-wasip1
cargo bench --target wasm32-wasip1-threads
WASM_HARNESS_ENGINE=sm cargo bench --target wasm32-wasip1   # pick SpiderMonkey

Engine can also be selected per-invocation by editing the script to pass --engine <name>, or by setting WASM_HARNESS_ENGINE in the environment.

Notes

• wasm-harness propagates a default whitelist of host env vars into the guest; the WASM_HARNESS_ENV_ prefix is the explicit escape hatch used here to set RAYON_NUM_THREADS without depending on whitelist contents.
• --inherit-env is available on wasm-harness if you'd rather forward every host env var instead of using the prefix.
• Bench code that builds its own rayon pool with an explicit num_threads() will ignore RAYON_NUM_THREADS — that's a per-bench concern, not a runner one.