fix(mqtt): bound QoS0 outbox and publish synchronously to stop heap exhaustion + packet drops

[agessaman]

Problem

Observer nodes running the MQTT bridge exhausted/fragmented internal heap under a
stalled-but-still-"connected" TLS uplink, and separately dropped ~20–30% of QoS0
packets even under light load. Two independent root causes, both in the QoS0 publish
path.

Root causes

1. Unbounded outbox. QoS0 packet publishes were forced into the esp-mqtt outbox
   (store=true) which has no size bound of its own — esp-mqtt frees entries only on
   send-ack or ~30s expiry. On a slow/half-open uplink, QoS0 frames accumulated on
   internal heap without limit.
2. ~1 msg/s drain ceiling. The esp-mqtt task sends only one QUEUED outbox item per
   loop iteration, and each iteration blocks up to MQTT_POLL_READ_TIMEOUT_MS (1s) on
   esp_transport_poll_read. With little inbound traffic that caps throughput at
   ~1 message/second per connection — below the ~1.2 pkt/s inflow — so the outbox pinned
   at its cap and the overflow was dropped. The poll timeout is a compile-time constant
   baked into the precompiled esp-mqtt lib, so the async drain can't be sped up on the
   Arduino/IDF 4.4 toolchain.

Changes

• Bound the outbox (PsychicMqttClient::setOutboxLimit() + a per-publish guard using
  esp_mqtt_client_get_outbox_size()): over the cap, publish() returns -2 and the
  bridge's existing retry/drop path applies backpressure. Caps: 16 KiB PSRAM / 8 KiB
  non-PSRAM. Portable across IDF 4.4 and 5; esp-mqtt's own outbox.limit config is not
  used (it doesn't reliably cover the QoS0 enqueue path).
• Publish QoS0 synchronously (esp_mqtt_client_publish(), async=false): writes
  straight to the socket, bypassing the drain ceiling entirely — QoS0 no longer touches
  the outbox. QoS1 status keeps the durable async/outbox + retransmit path. A stalled
  socket blocks only the Core-0 prio-1 MQTT task (mesh RX on Core 1 and the WiFi/TCP
  stack are unaffected), bounded by a new setNetworkTimeout() lowered to 2500 ms so a
  first stall fails fast and flips the slot to disconnected. The outbox cap remains as a
  dormant safety net.
• Diagnostics: per-slot publish ok/err counters. The periodic 30s heap/stats serial
  line is gated behind MQTT_MEMORY_DEBUG (off for production and plain MQTT_DEBUG
  builds; the heltec_v3 variant's explicit enable is now commented out). The same data is
  available on demand via a new get mqtt.stats CLI command.

On-target validation (~2h each)

• Station G2 (PSRAM, 4 brokers): Outbox=0 throughout, Max (largest free block)
  rock-stable 106–114 KB, ~9 publish errors over ~14k publishes (~0.06%, all at
  disconnect moments), 15 reconnects, zero crashes.
• Heltec V3 (non-PSRAM, 2 brokers): Outbox=0, 2 errors over ~8.5k publishes, zero
  crashes. Free stable.

Both boards: the original unbounded-outbox growth is gone and QoS0 drops fell to ~0.

Testing

• Builds clean: Station_G2_repeater_observer_mqtt and Heltec_v3_repeater_observer_mqtt
  (IDF 4.4 toolchain).
• get mqtt.stats confirmed working on-target.