macOS: all blueprints fail at startup — first coordinator→worker RPC (set_transport) times out on current main

[kumarganduri]

Environment: macOS 26.5 (Apple silicon), Python 3.12, install via uv,
transport default (zenoh). No robot attached.

Symptom: any dimos run / dimos --replay run fails ~2 min after module
deployment with:

TimeoutError: RPC call to 'RerunWebSocketServer/set_transport' timed out after 120.0 seconds

(The module name is whichever gets the first set_transport call.) All modules
deploy fine; zero Transport module=... log lines appear, so no
coordinator->worker RPC ever completes. Worker processes stay alive and log
nothing.

Repro:

uv run dimos --replay run unitree-go2-basic

Bisect evidence (same machine, same command):

repo state	result
29f35555 (mid-July)	works — transports assigned immediately, replay runs
7d301af5 (current main)	first set_transport RPC times out

Suspected cause: #3176 ("Replace RPC with native Zenoh RPC", DIM-1145) is
in the window, together with the eclipse-zenoh >=1.0 -> >=1.9 bump. The
new-main runs also print a zenoh warning the old base never showed, which
smells like peer discovery not connecting coordinator and workers on macOS:

WARN zenoh::net::routing::hat::peer: Gossip is disabled: cannot de-duplicate data traversing region gateways

Calling set_transport directly on a module instance in-process works fine,
so this looks confined to the RPC path between processes.

Happy to test candidate fixes or gather more logs from this machine — macOS
may just not be covered by CI for this path.