Terminate poisoned Playground PHP-WASM workers

[chubes4]

Summary

• overlay @wp-playground/cli@3.1.46 so worker v1/v2 terminate after a fatal WebAssembly.RuntimeError originating from php.wasm
• preserve the released CLI behavior for ordinary unhandled rejections
• cover the real worker-thread boundary for both worker variants

Root cause

The Playground CLI worker logs and swallows every unhandledRejection, behavior introduced by WordPress/wordpress-playground#2682. A fatal PHP-WASM runtime rejection poisons the worker, but the worker remains alive and the parent recipe request waits until its 25-minute timeout. The parent-process handling merged in #2011 cannot observe a rejection raised inside the worker thread.

This is a temporary downstream overlay while WordPress/wordpress-playground#4160 tracks the owning upstream repair.

Fixes #2010.

Verification

• npm ci
• npm run build
• npm run test:playground-worker-runtime-rejection
• npm run test:phpunit-runtime-rejection
• npm run test:playground-phpunit-readonly-cache-integration
• npm run test:php-wasm-extension-manifests

The worker regression verifies that both v1 and v2 workers survive an ordinary unhandled rejection, then exit nonzero and propagate the reproduced php.wasm.zif_mysqli_poll runtime rejection.

AI assistance

OpenAI GPT-5.6 Sol via OpenCode investigated the worker boundary, drafted the overlay and regression coverage, and ran the listed verification. Chris Huber reviewed and remains responsible for the change.