Pre-install the JavaScript RPC npm package before running tests

[timtebeek]

What's changed

test now depends on a warmJavaScriptRpcCache task that runs npx --yes --package=@openrewrite/rewrite@<version> rewrite-rpc once, using the version embedded in the resolved rewrite-javascript jar.

Why

CI has been failing intermittently for weeks with ~9 TypeScript tests reporting RPC process shut down early with exit code 1 (also 127 and 217, varying within a single run) — see run 30564235862, and previously 2026-07-15, 07-16, 07-19 and 07-28. It never reproduced locally.

JavaScriptRewriteRpc spawns npx --package=@openrewrite/rewrite@<version> rewrite-rpc, and the RPC process is shut down between test classes. Shimming npx on a runner showed four overlapping invocations in seven seconds, each triggering its own npm warn exec The following package was not found and will be installed, none of them reaching an exit:

=== 17:54:41.318702951 pid=3015 argv: --package=@openrewrite/rewrite@8.89.0-20260730-165737 rewrite-rpc
=== 17:54:44.022766233 pid=3053 argv: --package=@openrewrite/rewrite@8.89.0-20260730-165737 rewrite-rpc
=== 17:54:47.525195532 pid=3079 argv: --package=@openrewrite/rewrite@8.89.0-20260730-165737 rewrite-rpc
=== 17:54:48.681182317 pid=3097 argv: --package=@openrewrite/rewrite@8.89.0-20260730-165737 rewrite-rpc

An install takes ~3s, so on a cold cache these overlap in the same ~/.npm/_npx/<hash> directory and leave it half-written — one symptom being sh: 1: rewrite-rpc: Permission denied (exit 127) from a bin symlink whose target has not been chmod'ed yet. Locally the cache is always warm, so the install never happens and the race never opens.

Testing

On a clean ubuntu-latest runner, running the nine TypeScript test classes:

• without this change: all nine fail
• with this change: all nine pass

Follow-up

• The underlying race is in JavaScriptRewriteRpc/RewriteRpcProcessManager upstream — any consumer with TypeScript tests can hit it, and the discarded subprocess stderr is what made it undiagnosable for so long (cf. Surface RPC subprocess stderr instead of discarding it rewrite#8343). This change works around it here.