Omit SNR/RSSI instead of publishing hardcoded 12.5/-65 when no radio data

[ACETyr]

What

When the MQTT observer has no real radio measurement for a packet, omit the SNR/RSSI fields
instead of publishing the hardcoded constants 12.5 / -65.

src/helpers/MQTTMessageBuilder.cpp:

• buildPacketJSON() (the no-raw-data overload) previously passed literal 12.5f / -65 to
  buildPacketMessage() ("// SNR/RSSI - using reasonable default"). It now passes sentinels.
• buildPacketMessage() emits the SNR/RSSI fields only when the value is real (sentinel <= -900
  → omit). Real measurements are always within sane radio ranges and serialize exactly as before.

Why

buildPacketJSONFromRaw() carries the real per-packet SNR/RSSI, but the publish path falls back to
buildPacketJSON() whenever fresh raw radio data isn't available (no raw_data passed and the
global _last_raw_* snapshot older than 1 s — MQTTBridge.cpp). Because the MQTT publish runs on a
separate task that drains the queue asynchronously, that 1 s window is missed often.

The result is fabricated values indistinguishable from real readings. Measured against the public
CoreScope aggregator (200-packet samples per observer): the exact (SNR=12.5, RSSI=-65) pair appears
on ~3–20% of packets for every observer running this bridge (and 0% on a non-bridge observer),
silently polluting network-wide RF analytics (avgSnr/avgRssi//analytics/rf). Confirmed
end-to-end on a controlled node by correlating the serial RX log (real radio values) to the published
MQTT JSON by packet hash: published 12.5/-65 exactly where the radio reported real, different
values.

Omitting is honest about the gap — consumers can detect a missing field, but cannot detect a
fabricated reading.

Testing

• Compiles clean: pio run -e Heltec_v3_repeater_observer_mqtt → SUCCESS (ESP32-S3 image built).
• Localized JSON-builder change; buildPacketJSONFromRaw() (the real-data path) is unaffected.

Note

This only stops the fabrication. Reducing how often the fallback fires (e.g. carrying the
per-packet SNR/RSSI into QueuedPacket at reception so the publish path doesn't depend on the 1 s
_last_raw_* window) is a separate, optional robustness improvement.