Hardware bring-up verification (ESP32-S3 + Pico W)

[MichaelAdamGroberman]

The logic is covered by host tests (`pio test -e native`, 20/20) and CI proves the firmware builds for both boards — but on-device behavior has not been verified on physical hardware yet. This issue tracks running the hardware bring-up checklist once per board and reporting results.

Boards

• ESP32-S3
• Raspberry Pi Pico W (BLE is a documented stub — skip the BLE section; see roadmap)

Acceptance criteria (per the checklist)

1. Radio mutual exclusion (WiFi ⊕ BT)

• Boot in DEFAULT_MODE_WIFI → WiFi OK, no BLE advertising
• mode bt → serial shows WiFi OFF then BLE advertising… (ordered, never overlapping); BLE scanner sees the device only in BT mode
• mode wifi → BLE OFF then WiFi OK

2. Free heap reclaimed across a switch (the heap guarantee)

• Log free heap before/after a setMode switch
• Cycle WiFi→BT ~10× — free heap after each full cycle is stable (no monotonic decline = no leak; confirms stop hooks fully deinit, not just disconnect)

3. BLE transport (ESP32 only)

• Connect a BLE client, write token then read, get the {"temp_c":..,"uptime_s":..} notification; wrong token → silent drop

4. Optional WireGuard tunnel (WiFi mode)

• WG_CONFIG defined → tunnel reaches UP; route approved in Tailscale admin
• From another tailnet node: ping <device tunnel IP> succeeds
• From off-network: nmap -sU -p51820 <gateway> returns no WireGuard reply (scanner-silent)

5. Token-gated TCP (WiFi mode)

• printf 'TOKEN\nread\n' | nc <device> 4242 returns sensor JSON; wrong token closes with no data

Notes

• Flash via the bundled pio (Python ≤ 3.13): ~/.platformio/penv/bin/pio run -e <env> -d app/tailnet-sensor-node --target upload --target monitor
• On the ESP32-PICO-D4 the WiFi→BT teardown is the tightest heap case; the S3 has ample headroom.