Harden Dingtian: authenticated binary relay + disable unused channels

Lock down the relay device for the flat (no-VLAN) network.

Relay control:
- pulseOpen/setRelay now use the Dingtian BINARY protocol (:60000) with a
  relay password — the only relay option with auth (string :60001 has none, and
  is kept only for the read-only status query). Frame verified on hardware.

HardenableDevice capability (driver harden()):
- set a random relay_pw (1-9999); disable unused channels (rs485/can/tcp x2/mqtt
  -> p:255), keeping UDP1 binary (control) + UDP2 string (status).
- write-verified (device reboots on apply).

Assign/Save flow now does: fix preconditions -> harden -> set up input push;
the relay password is stored in lane_devices so the runtime device can command
the relay.

DELIBERATELY NOT touching the device's HTTP CGI session check (session_en):
enabling it on this firmware breaks the config-READ API (ECONNRESET) and locked
the backend out — required a factory reset to recover. The open CGI API is
accepted as flat-network reality; the signed event log is the real guarantee.

Verified end to end on hardware: assign hardens + configures the device, config
API stays reachable, pulseOpen with the stored password fires the relay, without
it is rejected. wiki: device-input-flow + dingtian-relay updated.

wiki/concepts/device-input-flow.md

@@ -33,20 +33,39 @@ car arrives → driver presses button (input I_N, dry contact to GND)
 
 ## Trust model (important — flat network, no VLAN)
 
-The relay-control direction (host → device) is **unauthenticated UDP**, and the site is a **flat
-network with no VLAN** ([[network-isolation]] is not yet enforceable here). So we do **not** trust
-the device or the network. Instead:
+The site is a **flat network with no VLAN** ([[network-isolation]] is not yet enforceable here),
+so we do **not** trust the device or the network. Both directions now have defence-in-depth, but
+neither is the real boundary:
 
-- **Every barrier open is a host decision, recorded as a signed event BEFORE the relay fires**
-  ([[append-only-event-chain]]). If anyone opens the relay out-of-band (which the flat network
-  allows), there is **no matching signed event → a detectable anomaly**. The anti-fraud guarantee
-  is the **signed log**, not device/network auth.
-- The inbound push endpoint is intentionally **not behind the SPA's cookie/CSRF auth** (it's a
-  machine call from the device). It is guarded by **HTTP Digest auth** + a **source-IP allowlist**
-  (defence-in-depth), but these are *not* the security boundary.
+- **Relay control (host → device)** — UDP, now via the Dingtian **binary protocol on :60000 with a
+  `relay_pw`** (the only authenticated relay option; the string protocol has none). Set on the
+  device + stored in `lane_devices` by the harden step (below).
+- **Input push (device → host)** — guarded by **HTTP Digest auth** + a **source-IP allowlist**.
+- **The real guarantee is the signed log:** every barrier open is a host decision, recorded as a
+  signed event BEFORE the relay fires ([[append-only-event-chain]]). An out-of-band open (which a
+  flat network allows) has **no matching signed event → a detectable anomaly**. Device/network
+  auth is just speed bumps; both are plaintext over a sniffable network.
 - This sharpens under the [[autonomous-direction|unmanned]] roadmap: with no operator, tamper
   detection via the signed log matters more than perimeter auth.
 
+## Device hardening (on assign)
+
+The assign/Save step configures the device end-to-end (admin never touches the device web UI):
+fix preconditions (disable `input_link_relay`) → **harden** → set up input push. The `harden`
+capability ([[device-registry|HardenableDevice]]):
+
+- **Sets a random `relay_pw`** (1–9999) so binary relay commands need it; stores it in
+  `lane_devices` so the backend can keep commanding the relay.
+- **Disables unused protocol channels** (rs485, can, tcp×2, mqtt → `p:255`), keeping only UDP1
+  binary (relay control) + UDP2 string (status read) — fewer open doors.
+
+> **⚠️ Lesson (the hard way):** do **NOT** enable the device's HTTP CGI session check
+> (`session_en`). On this firmware (DT-R004) it makes the config-**read** API drop connections
+> (`ECONNRESET`), locking the backend out of the very API it depends on — it required a **factory
+> reset** to recover. The harden step deliberately leaves `session_en` off. The CGI config API
+> being open is accepted as part of the flat-network reality (the signed log is the guarantee);
+> the proper fix is network isolation, not this fragile device feature.
+
 ## Push authentication — Digest (decided by hardware testing)
 
 The secret must not be in the URL (sniffable, logged) and the password must not cross the wire in