MQTT publish ring buffer fills and drops continuously with MQTT disabled and no broker configured; followed by panic crash (v3.2.0-rc.3, ESP32-S3)

[geertvanvaerenbergh-lgtm]

Setup

• Gateway: ESP32-S3, firmware v3.2.0-rc.3
• Charger: Wallbox Pulsar Plus (SN 343025), recently auto-updated charger firmware
• "HA MQTT Discovery" toggle: Off ("I use the HACS Integration")
• MQTT broker host field: empty (never configured; only default port 1883 present)
• Polling: HACS integration over HTTP, 60 s interval
• BLE link healthy throughout: -39 to -55 dBm, 0 BLE reconnects

Observed behavior

1. The serial/telnet log fills continuously with:
   [BLE] pending MQTT pub ring full, dropping met=<r_lse|r_not|g_alo|g_ecos|g_psh|g_phsw|g_tzn|r_dca|g_halocfg|...>
   This occurs on nearly every BLE poll cycle, indefinitely.
2. This happens although MQTT Discovery is Off and no broker is configured at all. Expected: with MQTT disabled/unconfigured, BLE results should not be queued for MQTT publish in the first place.
3. The gateway subsequently suffered a panic crash ("Last boot: panic (crash) — 1 bad boot on this firmware" on the Info page). Prior boot had run ~14.5 h (NVS: "WIFI at +14h37m of that boot, down 0s"), so WiFi was stable; heap looked healthy earlier (≈105 KB free, min watermark ≈65 KB). I can't prove the ring buffer behavior caused the panic, but it is the only persistent anomaly in the logs.

Expected behavior

• No MQTT publish queuing when MQTT is disabled / no broker configured (or at minimum, a one-time notice instead of continuous per-message drops).
• No panic.

Additional info

Happy to provide the full log capture, /info diagnostics, or run a debug build.