F-23: WiThrottle heartbeat (JMRI dead-man stop) never armed #26

Severity: HIGH · Effort: Small · Batch: 2 · Status: Code done in #21 (68452a5), awaiting bench verification

Found in the second full review of the firmware. Recorded with the rest of the pass (F-19…F-44) in docs/REVIEW_REMEDIATION_PLAN_2.md, which is where its status is kept.

Description

WiThrottleClient::connect() never sends *+, and nothing sends a periodic *. The repo's own
protocol notes call the heartbeat "CRITICAL". Without it, JMRI never e-stops this device's locos
when the device crashes, reboots, or goes half-open on WiFi — the dead-man switch is simply
off. The device also sends HESP32-S3, where the protocol expects HU<unique-id>.

Plan

1. Send HU<MAC> (a stable per-device id) and N<name> on connect.
2. Parse the server's *<seconds> announcement; if non-zero, send *+ and then * at half
   that interval. Sent from the receive task, which wakes at least once a second anyway, so
   the heartbeat needs no task of its own and ends with the session.

Files: WiThrottleClient.h/.cpp

Acceptance Criteria

• *+ is sent once per session when the server announces a non-zero interval.
• * is sent at half the announced interval for the life of the session, from the
  receive task (which now wakes every second).
• HU<station MAC> replaces the malformed HESP32-S3.
• Unit tests test_withrottle_heartbeat_announcement_arms_monitoring and
  test_withrottle_malformed_heartbeat_is_ignored pass on the board (they compile).
• Bench: pull the ESP's power with a loco running; JMRI e-stops it within the interval.