Feature: Add live channel-health metrics to device displays

[usrflo]

Motivation

The statistics a MeshCore device reports to the app are cumulative lifetime totals - they say nothing about the current situation. Channel stats over a short time window let a user quickly check the health of the mesh at their location.

This adds three small, windowed channel-health metrics rendered directly on the device display (companion ui-new/ui-tiny and simple_repeater) and exposed via the stats-radio JSON.

The three metrics

Line	Question it answers	Window	What feeds it
CH free	How much airtime is free?	~5 s	own TX + mid-receive + RSSI above the quiet-channel reference (see below) with a fixed +15 dB margin, sampled at least every 50 ms
RX ready	How much of the time was the radio actually in RX mode?	~5 s	isInRecvMode() sampling in the radio loop - catches duty-cycle deafness, blocking code, sleep misconfig
RX quality	Of the packets that should decode, how many did?	~10 min	good vs total decodes; CRC failures only counted when SNR ≥ per-SF threshold + 3 dB guard, and reported as 0 % while reception is jammed dead (see below)

All three render as uniform health bars. A radio that cannot measure them (e.g. ESP-NOW builds) reports no data, and the UI shows "-- %" with a dimmed bar instead of a false "all healthy".

Sustained interference

The channel-busy verdict needs a reference for "quiet". Using the adapted noise floor directly has a failure mode: a floor that tracks a sustained interferer hides that interferer behind its own margin, so a jammed channel reads as free. CH free therefore measures against a quiet-channel reference: the 10th percentile of the last 64 published noise-floor values (covering several minutes), absolutely capped at −100 dBm. A sustained broadband interferer keeps CH free near 0 % for as long as it is present — even for hours — and the reference recovers on its own within minutes once the channel is quiet again. During warm-up (first 8 blocks after boot) the current noise floor is used as reference.

RX quality cannot drop on its own when reception dies: a jammed channel produces no decode attempts at all, so the windowed ratio would freeze at its last healthy value. It therefore reports 0 % when ambient energy stayed above the quiet reference for ≥ 80 % of the last 5 s and no decode attempt occurred for 2 minutes — meaning reception is dead, not merely "no traffic". While packets still decode through the interference, the honest good/total ratio is shown; genuine quiet stays "-- %".

This is display-only: LBT (interference_threshold, CAD, resend deferral) keeps using the adapted noise floor. The verdicts reuse the existing ≥ 50 ms RSSI poll — no extra SPI traffic; the percentile is recomputed once per ~3.2 s calibration block.

Field observation (site with an intermittent broadband interferer ~37 dB above the −113 dBm ambient): while present, CH free reads 0 % and RX quality falls to 0 % within ~2 min, with the noise-floor row showing −76; once it disappears, all three rows recover within minutes.

stats-radio JSON

Short keys util / deaf / good / tot alongside the existing chan_util_pct / rx_deaf_pct / rx_good / rx_total (the reply buffer is 160 bytes).

Testing

• mcsim end-to-end: utilization/deafness rise with traffic, good == tot in clean runs, JSON reply ~106 B
• Builds: Heltec_v3_repeater, RAK_4631_repeater, LilyGo_T-Echo_Card, Heltec_t1 companion, WioTrackerL1 companion
• Hardware field testing on a WioTrackerL1 companion (observation above) and Heltec V3