so yea, i'll keep working on https://github.com/pantsbuild/pants/issues/5169 ... after the scheduler lock removal the hang moved from

acquire

on the scheduler lock to a hang inside

run_and_return_roots

(ie, on the graph lock)