daikin_madoka: BLE watchdog + multi-client gattc/gap filters

[pridmen]

Three orthogonal hardening fixes for ESP32 setups that pair more than one BRC1H and need to survive Wi-Fi flaps.

1. BLE watchdog

After a Wi-Fi flap (or a stray bluedroid race) the BLE stack occasionally reports ESTABLISHED but the pulse never replies again — parse_cb_() stops, current_temperature stays NAN, and HA's commands disappear into the void. Only a physical power-cycle of the ESP32 recovers.

Track last_response_ms_; if it exceeds 5 minutes while node_state is ESTABLISHED, issue a per-channel disconnect (the other BRC1H pair is left alone). After 3 consecutive failures, fall back to App.safe_reboot() as a safety net.

• parse_cb_() refreshes the timestamp on every successful reply.
• REG_FOR_NOTIFY_EVT refreshes the timestamp on (re)connect so a long disconnect interval does not immediately trip the watchdog.
• The watchdog only runs once a reply has been seen at all (last_response_ms_ > 0), so a fresh flash with an unpaired pulse does not reboot-loop.

2. Multi-client gattc_if filter

Recent ESPHome versions dispatch the same gattc_event_handler to every BLEClientNode registered against the same esp32_ble interface, regardless of which BLE connection the callback actually belongs to. Without filtering, a DISCONNECT_EVT from the OTHER BRC1H wipes our current_temperature/target_temperature to NAN — observed on dual-pair setups where one pulse stays connected and the other is in a reconnect loop.

Filter at the top of the handler by gattc_if.

3. Multi-client remote_bda filter (GAP variant)

Same broadcast issue affects gap_event_handler too. An AUTH_CMPL_EVT from the OTHER BRC1H reaches our handler with a remote address that doesn't match this->parent_->get_remote_bda(), drives try_register_notifications_() against the wrong remote, and logs noisy No control service found at device, not a Daikin Madoka..? warnings. GAP events don't carry a gattc_if, so filter by remote_bda extracted from param->ble_security.{ble_req,auth_cmpl}.bd_addr.

Test plan

• Two paired BRC1H pulses on a single ESP32 — watchdog triggered correctly after artificial Wi-Fi flap, per-channel disconnect cleanly restored BLE link.
• gattc_if filter — DISCONNECT events from one BRC1H no longer wipe current_temperature of the other one (reproducible regression on ESPHome 2026.5).
• remote_bda GAP filter — eliminates No control service found log noise during reconnect of the second pair.
• Standalone single-pulse setup — no behaviour change (watchdog never trips while polling returns within 5 min).

🤖 Generated with Claude Code

[github-actions]

To use the changes from this PR as an external component, add the following to your ESPHome configuration YAML file:

external_components:
  - source: github://Petapton/esphome@pull/16/head
    components: [daikin_madoka]
    refresh: 1h

---

(Added by the PR bot)

[Petapton]

Can't reproduce. Please at least provide logs.