dev-stack.sh down leaves the dApp dev server holding port 3012

[lmcorbalan]

Description

down stops the containers and the pidfile-tracked processes, but the Vite dev server survives
and keeps listening on 3012. Its fallback pkill -f "vite --host localhost --port 3012" never
matches: app:dev passes its flags after --, so the real argv is
vite -- --host localhost --port 3012 --strictPort. stop_pidfile misses it too, because
pkill -P kills only direct children and vite is a grandchild of the recorded pid.

Steps to reproduce

1. Start Docker, then run ./scripts/dev-stack.sh up
2. Wait for "Stack is up" and confirm http://localhost:3012 answers
3. Run ./scripts/dev-stack.sh down
4. Run lsof -nP -iTCP:3012 -sTCP:LISTEN and see a node process still listening

Expected vs actual behavior

Expected: down frees 3010-3013 and its own port report prints (all free).
Actual: a node process still holds 3012, down's own report lists it, and the next up
warns "Port 3012 already in use" and skips the dApp.

Environment

Observed 2026-09-08 while running the stack on feat/185-dev-stack-no-dpm
The pkill pattern dates from 1dfa3ce (2026-06-11), unrelated to the token-forge migration

Additional context

Killing the recorded pid's whole process group would reach the grandchild; matching the real
argv would fix the fallback.