RM: configurable CEM endpoint, stable structural IDs, richer dashboard

[martijnversluis]

Three focused fixes to the rm-mqtt-ws-bridge reference implementation. Each is a separate commit so it can be reviewed on its own.

1. Route the RM to a configurable CEM endpoint

Today the RM in the three-panel demo is hard-wired to the bundled stand-in in rm-mqtt-ws-bridge/cem/. This makes it awkward to interoperate with cem-reference-1 or any external CEM without editing the compose file.

The change turns CEM_BASE_URL into a proper knob:

• External CEM — CEM_BASE_URL=wss://your-cem.example/ws ./demo.sh routes the RM there and skips the bundled stand-in.
• Reference CEM — with no CEM_BASE_URL, demo.sh probes cem-reference-1 on localhost:8003/health and, when it answers, wires the RM to ws://host.docker.internal:8003/ws. Cross-container-to-host works through the newly added extra_hosts: host.docker.internal:host-gateway on the RM service.
• Simulation — otherwise it falls back to the bundled cem stand-in, now guarded by a bundled-cem compose profile so it only starts in this mode. depends_on: cem on both rm and demo uses required: false so the graph stays valid without it.
• MOSQUITTO_HOST_PORT and RESOURCE_ID were also promoted to passthroughs so the defaults can be overridden without editing files.

The three-panel demo screen (demo/index.html) was hard-coding ws://localhost:9000/s2 and always framing the CEM stand-in iframe. It now fetches /traffic from the RM at page-load (and every 2 s), fills the address bar with the live cem_endpoint, and, when that URL points outside the compose network, replaces the CEM iframe with an "S2 CEM external" info tile that shows the endpoint and the link state. When the bundled stand-in is active the iframe is restored, so the original click-through walkthrough still works. TrafficController sends Access-Control-Allow-Origin: * so the demo panel can read /traffic cross-origin.

2. Keep FRBC system description IDs stable across reconnects

FRBCSystemDescriptionFactory generated the actuator id, both operation mode ids, and both transition ids with SecureRandom.uuid. Every reconnect (or RM restart) produced a fresh set, so any FRBC.Instruction a CEM had already persisted referenced actuator/operation_mode ids that no longer existed in the new system description. On the wepositive CEM this shows up as a red warning triangle next to those fields on stored instructions.

The structural ids are now derived deterministically with Digest::UUID.uuid_v5 from a fixed namespace and <resource.s2_identifier>/<role>. Same resource + same role → same UUID, across reconnects and restarts, with no persistent state to maintain. Only message_id stays random per emission, as the S2 spec requires.

3. Fold reception statuses and raw bodies into the RM dashboard feed

The RM dashboard feed had four rough edges:

• ReceptionStatus was rendered as its own row, doubling the feed length and hiding which message it acked.
• Handshake / HandshakeResponse / SelectControlType / ResourceManagerDetails / FRBC.SystemDescription rows had no direction indicator, so RM→CEM and CEM→RM looked identical.
• The raw S2 JSON was not viewable; the MQTT panel already offered "> raw message".
• The InstructionStatusUpdate that the RM sends back after processing an FRBC.Instruction was silently dropped by the render filter.

The change:

• TrafficLog now stores message_id, subject_message_id, status, and the parsed body per entry. S2MessageLogger and MQTTMessageLogger extract those fields from the payload.
• ReceptionStatus is no longer its own row. Every S2 message row carries a small badge: green ✓ (OK), red ! (any other status), grey … (waiting). Clicking the badge opens the raw ReceptionStatus JSON inline in a green-tinted block.
• Clicking anywhere else on a message row toggles the raw JSON body of that message, matching the "> raw message" affordance of the MQTT panel. Open/closed state survives the 1.2 s poll cycle via a stable id hashed from peer/direction/timestamp/message_id/subject_message_id.
• Setup rows now carry a direction arrow so RM→CEM (←, toward the CEM column) and CEM→RM (→) are distinguishable at a glance, matching the existing arrow convention on report and command rows.
• The FWD map gained instruction_update → InstructionStatusUpdate, so the reverse-path acknowledgement now surfaces on its own row (with the CEM's ack badge) instead of being filtered out.

[jorritn]

Cool. Really nice improvements @martijnversluis

One thing worth mentioning: it is not wrong to use reuse Ids in the FRBC system description but from the S2 pov also not required: a CEM should not require the Ids to be stable over multiple sessions. Every session is fully self contained and state is not supposed to be shared over multiple session. Nevertheless, for debugging purposes and system insights it can be helpful to use the same Ids.