readPid() trusted posix_kill($pid, 0), which only proves some process
is alive at that PID. When storage/torque.pid survives a container
restart on a bind mount, the PID number is routinely recycled to an
unrelated process, so torque:start refused to boot ("already running"),
torque:status reported a phantom master, and reload/flush/monitor were
misled the same way.

readPid() now also checks /proc/<pid>/cmdline for "torque:start" before
trusting the PID file. A recycled or dead PID is treated as stale and
unlinked; startup proceeds and the existing pgrep orphan sweep still
handles genuine orphans. Platforms without /proc (macOS) keep the
liveness-only behaviour.