Fix SpoolBuddy scale calibration lost after reboot

  _get_mac_id() used Path.iterdir() to find a network interface for the
  device ID, but iteration order is non-deterministic. On different boots
  it could pick eth0 (MAC …3100) or wlan0 (MAC …3102), producing a
  different device_id each time. Since calibration is stored per device ID
  in the backend DB, a new ID meant registering as an uncalibrated device.

  Sort interfaces alphabetically so the same one is always selected.

CHANGELOG.md

@@ -27,6 +27,7 @@ All notable changes to Bambuddy will be documented in this file.
 - **A1 Mini Shows "Unknown" Status After MQTT Payload Decode Failure** ([#549](https://github.com/maziggy/bambuddy/issues/549)) — Some printer firmware versions (observed on A1 Mini 01.07.02.00) occasionally send MQTT payloads containing non-UTF-8 bytes. The `_on_message` handler called `msg.payload.decode()` (strict UTF-8), and the resulting `UnicodeDecodeError` was not caught — only `json.JSONDecodeError` was handled. The entire message was silently dropped, causing printer status to show "unknown", temperatures to read 0°C, and AMS data to disappear. Now catches `UnicodeDecodeError` and falls back to `decode(errors="replace")`, which substitutes invalid bytes with U+FFFD while keeping the JSON structure intact. Logs a warning for diagnostics.
 - **H2C Dual Nozzle Variant (O1C2) Not Recognized** ([#489](https://github.com/maziggy/bambuddy/issues/489)) — The H2C dual nozzle variant reports model code `O1C2` via MQTT, but only `O1C` was in the recognized model maps. This caused the camera to use the wrong protocol (chamber image on port 6000 instead of RTSP on port 322) — the printer immediately closed the connection, producing a reconnect loop. Also affected model display names, chamber temperature support detection, linear rail classification, and virtual printer model mapping. Added `O1C2` to all model ID maps across backend and frontend.
 - **Support Package Leaks Full Subnet IPs and Misdetects Docker Network Mode** — Three support package fixes. First, the network section included full subnet addresses (e.g., `192.168.192.0/24`); now masks the first two octets (`x.x.192.0/24`). Second, `network_mode_hint` used `len(interfaces) > 2` which always reported "bridge" on single-NIC hosts even with `network_mode: host`, because `get_network_interfaces()` excludes Docker infrastructure interfaces. Now checks for the presence of Docker interfaces (`docker0`, `br-*`, `veth*`) via `socket.if_nameindex()` — these are only visible when the container shares the host network namespace. Third, `developer_mode` was still null for most users because the MQTT `fun` field was only parsed inside the `print` key; some firmware versions send it at the top level of the payload. Now also checks top-level `fun`. Also added a `virtual_printers` section with mode, model, enabled/running status, and pending file count for each configured virtual printer.
+- **SpoolBuddy Scale Calibration Lost After Reboot** — The SpoolBuddy daemon generated its device ID from the MAC address of whichever network interface `Path.iterdir()` returned first, but filesystem iteration order is non-deterministic. On different boots, the daemon could pick `eth0` (MAC ending `3100`) or `wlan0` (MAC ending `3102`), producing a different `device_id` each time. Since calibration values (`tare_offset`, `calibration_factor`) are stored per device ID in the backend database, a new ID meant registering as a brand-new uncalibrated device. Fixed by sorting network interfaces alphabetically before selection, ensuring the same interface (and thus the same device ID) is always chosen.
 - **SpoolBuddy NFC Reader Fails to Detect Tags** — The PN5180 NFC reader had two polling issues. First, each `activate_type_a()` call that returned `None` (no tag) corrupted the PN5180 transceive state — subsequent calls silently failed even when a tag was physically present, making it impossible to detect tags placed after startup (only tags already on the reader during init were detected). Fixed by performing a full hardware reset (RST pin toggle + RF re-init, ~240ms) before every idle poll, giving a ~1.8 Hz effective poll rate. Second, after a successful SELECT the card stayed in ACTIVE state and ignored subsequent WUPA/REQA, causing false "tag removed" events after ~1 second. Fixed with a light RF off/on cycle (13ms) before each poll when a tag is present, resetting the card to IDLE for re-selection. Also added error-based auto-recovery (full hardware reset after 10 consecutive poll exceptions), periodic status logging every 60 seconds, and accurate heartbeat reporting of NFC/scale health.
 
 ### Improved

spoolbuddy/daemon/config.py

@@ -61,9 +61,14 @@ class Config:
 
 
 def _get_mac_id() -> str:
-    """Generate a device ID from the primary network interface MAC address."""
+    """Generate a stable device ID from the primary network interface MAC address.
+
+    Interfaces are sorted by name so the same interface is always picked
+    regardless of filesystem iteration order (eth0 before wlan0, etc.).
+    """
     try:
-        for iface in Path("/sys/class/net").iterdir():
+        ifaces = sorted(Path("/sys/class/net").iterdir(), key=lambda p: p.name)
+        for iface in ifaces:
             if iface.name == "lo":
                 continue
             addr_file = iface / "address"