Add RSSI-based listen-before-talk

[jirogit]

Motivation

MeshCore has two pre-transmit channel checks today, and neither can express the requirement this PR addresses.

cad uses the radio's hardware Channel Activity Detection, which looks for a LoRa preamble. In a band shared with equipment using other modulations, that is exactly the traffic it cannot see.

int.thresh does sample RSSI, but relative to the tracked noise floor, and only once. A deployment that has to hold to an absolute dBm threshold, sense the channel for a defined window rather than instantaneously, or stay quiet for a fixed period after each transmit, has no way to say so.

This PR adds RSSI-based listen-before-talk: modulation-agnostic energy detection with an absolute threshold and a configurable sensing window, plus three transmit-shaping limits that go with it.

What's new

set rssi.lbt on|off
get rssi.lbt -> "> on" / "> off"

set rssi.lbt.params <thr_dBm>,<sense_ms>,<maxwait_ms>,<txmax_ms>,<pause_ms>
get rssi.lbt.params -> "> -80,5,0,4000,50"

get rssi.lbt.params reports the values in the same order set takes them and with the same separator, so its output can be pasted straight back into set — following the existing get radio / set radio pair. Its arguments are:

argument	default	meaning
thr_dBm	-80	absolute RSSI threshold; a sample above this marks the channel busy
sense_ms	5	how long RSSI is sampled for. 0 takes a single instantaneous reading
maxwait_ms	0	how long a busy channel may hold off a pending transmit before it is forced out. 0 = unlimited
txmax_ms	4000	max airtime of one transmit; a longer packet is dropped rather than sent. 0 = unlimited
pause_ms	50	quiet period enforced after each transmit

Throughout, 0 means "unlimited". The defaults are chosen so that a single set rssi.lbt on yields a usable configuration without touching rssi.lbt.params.

Behaviour when disabled

This is the first thing worth checking, so to state it plainly:

• While rssi.lbt is off — the default — every one of the rssi.lbt.params values is ignored. The sensing loop, the post-transmit pause, the airtime cap and the max-wait are each gated on the enable flag, not merely on their own value being zero. A node that never issues these commands transmits exactly as it does on dev.
• getCADFailMaxDuration() returns the original hard-coded 4000ms whenever rssi.lbt is off, whether or not cad is on.
• The legacy binary prefs loader is intentionally not extended. These fields cannot exist in a file written by a build that predates them, so an upgraded node simply starts ith the defaults above.

Design notes

Independent of CAD. The RSSI check is added as a third branch of RadioLibWrapper::isChannelActive(), next to int.thresh and _cad_enabled. The two existing branches are unchanged and the return value stays a plain bool, so isReceiving() remains the single gate Dispatcher::checkSend() consults. Any combination of the three may be active.

Max-wait. rssi.lbt's maxwait_ms and CAD's fixed 4000ms both bound the same busy state in checkSend(). With rssi.lbt off the function returns 4000 unchanged. With it on, the stricter — longer — wait applies, so CAD's fixed value never shortens a wait that rssi.lbt asked to be longer. Since 0 means "never force the transmit", it is normalised to the largest representable duration before any comparison, so it does not lose to a finite value.

The effective value is derived on every call and never written back to prefs, so a computed number cannot end up stored and disagree with what the CLI reports.

Airtime cap. Enforced immediately before startSendRaw(), reusing the existing send-failure cleanup path. There is no counter for dropped packets; the drop is visible only through MESH_DEBUG_PRINTLN, so release builds stay silent.

One implementation, four node types. The prefs live in CommonRadioPrefs, which both NodePrefs::RadioPrefs implementations derive from, so companion, repeater, room server and sensor are all covered without per-target code.

Commits

The PR is one change but is split into four commits that can be taken separately if review stalls on any one of them:

1. The RSSI sensing mechanism and post-transmit pause in RadioLibWrapper. Inert on its own; nothing enables it yet.
2. The prefs and CLI wiring.
3. The airtime cap and the max-wait handling.
4. Documentation.

Testing

Firmware builds and basic on-device smoke testing:

• M5Stack Unit C6L (ESP32-C6 / SX1262) — repeater
• Seeed T1000-E (nRF52840 / LR1110) — repeater
• RAK WisMesh Tag (nRF52840 / SX1262) — companion, over both BLE and USB

Nodes come up, advertise and forward packets as before, and with rssi.lbt left at its default of off I saw no behaviour change from current dev. This was smoke testing rather than a measured comparison; I have not run the mechanisms under sustained traffic.

Open question

pause_ms is currently a blocking delay() in onSendFinished(). At the default of 50ms that seemed acceptable, but I am happy to make it non-blocking if you prefer.