fix(bridge): retake the transport when its owner dies later

[Sycatle]

Reclaiming a stale lock at startup was already right. The mistake was treating that one decision as final.

A daemon that lost the race stayed degraded for its whole life, even once the winner exited. health went on reporting owns: false and naming a dead PID, and the conflict message told the operator to close a session that was already closed.

How it surfaced

A second session hit it today: it removed the stale lock file and nothing changed, because the losing daemon had read that file once at boot and kept the verdict in memory. It ended up killing its own daemon and losing the transport until a human reconnected it — a self-inflicted outage caused entirely by the tool describing a competitor that no longer existed.

Same family as the three map-reading bugs fixed earlier today: not wrong about the bytes, wrong about what they designate.

The fix

status() re-attempts the acquisition whenever it does not hold the lock.

It deliberately does not test the recorded PID first. The acquisition is atomic (wx) and already refuses a live owner, so retrying blind is safe — and it covers what a liveness test misses: an owner that released the file cleanly while its process is still around. My first attempt did gate on isAlive, and the new test caught it.

Two negative controls, both watched red

• Never reclaim → the late-death test fails.
• Reclaim unconditionally → three tests fail, including the one asserting a live owner keeps the directory. That is the two-daemon failure this lock exists to prevent, and the reason the retry has to go through acquireTransportLock rather than around it.

132 tests.