Vouch request: fix orphaned trigger-dev-run-worker processes after SIGKILL #3478

Why do you want to contribute?

I'd like to fix issue #2909: orphaned trigger-dev-run-worker processes that accumulate and consume significant CPU when the CLI is killed ungracefully (e.g. via pnpm's SIGKILL).

I traced the bug through DevSupervisor, TaskRunProcessPool, and the existing watchdog introduced in #3191. The watchdog correctly handles server-side run cancellation, but doesn't kill the OS-level worker processes on the user's machine. My fix extends the watchdog to read worker PIDs from active-runs.json and SIGTERM -> SIGKILL them when the parent CLI dies.

I've tested this end-to-end against a local self-hosted instance, SIGKILL the CLI, wait ~5 seconds, and all worker processes are dead.

The change touches three files: TaskRunProcessPool (adds getAllPids()), DevSupervisor (writes workerPids to active-runs.json with a 2s refresh), and devWatchdog (kills those PIDs on parent death).

Prior contributions or relevant experience

I'm a backend engineer with 5+ years of experience building distributed systems in Node.js/TypeScript. My stack closely mirrors Trigger.dev's internals, NestJS, TypeScript, PostgreSQL, Redis, BullMQ, and queue-based job orchestration. I have a published research preprint on a Multi-Pipeline Outbox Pattern as an alternative to Kafka for moderate-scale systems (DOI: 10.21203/rs.3.rs-8328306/v1). I've also contributed to Debian's open-source projects through the Debian JavaScript Team (Outreachy internship, DebConf 2022 presenter).

GitHub: https://github.com/thegodtune