Function wait_for_worker_ready 

pub fn wait_for_worker_ready(
    worker: &mut PayloadHandle,
    worker_pid: u32,
    timeout: Duration,
    role: &str,
    exit_code_legend: &str,
) -> Result<()>

Poll for the worker’s ready marker with a deadline, returning early if the worker exits before writing the marker or after writing but before the caller’s subsequent dispatch.

Event-driven via inotify on the marker’s parent directory (IN_CREATE | IN_MOVED_TO). The wait wakes on the actual file- create edge — kernel-scheduling-tick latency — instead of a 10 ms poll tail. inotify is set up BEFORE the initial existence probe so a marker that lands between probe and watch-add still fires the watch on a subsequent edge; the loop body re-checks existence on every wake so a stale watch edge from an unrelated file in the same directory doesn’t false-fire.

The caller supplies role (e.g. "worker", "churn worker") and exit_code_legend (a variant-specific decoder for the worker-binary exit codes the caller wants printed in the error message). Worker cleanup on timeout happens via PayloadHandle::drop when the caller’s Result error propagates — calling PayloadHandle::kill(self) here would take the handle by value, which we can’t do behind an &mut borrow.

Consolidates what used to be two near-identical 20-line poll loops in tests/jemalloc_probe_tests.rs — a rename of the marker path, a change in poll interval, or a new early-exit shape now edits one site instead of two.