Recurring RESET_WATCHDOG causing full Z2M restarts every ~30-90s on HA Yellow (ember, firmware 7.4.4)

What happened?

Zigbee2MQTT restarts completely every 30-90 seconds due to repeated
RESET_WATCHDOG errors from the coordinator. This happens continuously
during normal operation (not just at startup), causing devices to
intermittently become unavailable and requiring manual add-on restarts
when the built-in watchdog stops recovering automatically.
Log pattern (repeats continuously):
[time] error: zh:ember:uart:ash: Received ERROR from adapter while connecting, with code=RESET_WATCHDOG.
[time] info: z2m: Starting Zigbee2MQTT version 2.13.0 (commit #unknown)
Starting Zigbee2MQTT without watchdog.
[time] info: z2m: zigbee-herdsman started (resumed)
Alongside this, we also see frequent network/route errors
(ROUTE_ERROR_MANY_TO_ONE_ROUTE_FAILURE, ROUTE_ERROR_SOURCE_ROUTE_FAILURE,
ROUTE_ERROR_NON_TREE_LINK_FAILURE) and occasional ZCL command delivery
timeouts to various devices during/after each restart cycle.

What did you expect to happen?

Stable coordinator connection without recurring watchdog resets.

How to reproduce it (minimal and precise)

No specific trigger identified. Occurs intermittently during normal
operation with ~87 automations and a Zigbee network of moderate size
(mix of routers and end devices, mains-powered and battery).
Already tried:
Setting serial.rtscts: false explicitly — no improvement, watchdog
errors continued at the same frequency.
Reverted to serial.rtscts: true (the documented default for ember)
— no improvement either.
Checked for a single problematic device by correlating route-error
short addresses in the logs — two devices appeared more often than
others (one end device, one router/Tuya thermostat), but both are
fixed/mains-powered devices that haven't moved or lost power recently,
so we don't believe either is the root cause; more likely a symptom
of the coordinator reset than the cause.
Checked host system logs for OOM kills — found unrelated OOM events
(a different container was being killed) at a different time of day,
not correlated with the watchdog reset timestamps.

Zigbee2MQTT version

2.13.0

Adapter firmware version

7.4.4 [GA] (EZSP 13, EmberZNet)

Adapter

Home Assistant Yellow (internal Silicon Labs radio, ember driver, /dev/ttyAMA1)

Setup

Host: Home Assistant Yellow (CM4, 4GB RAM)
Add-on watchdog is enabled (watchdog: true) and does trigger
automatic restarts most of the time, but has occasionally required
a manual restart when it stopped recovering on its own.
We're hesitant to attempt a firmware downgrade to 7.4.1 (mentioned
in older threads like #21462) since that thread's RESET_WATCHDOG
symptom was specifically at startup, whereas ours occurs continuously
during normal operation — and current HA Yellow radio firmware
update prompts suggest 7.4.4 is the currently recommended version,
so we wanted to ask before downgrading against that recommendation.
Happy to provide a full debug log capture or the Ember Helper
(nerivec.github.io/z2m-ember-helper) diagnostic if useful.

Device database.db entry

No response

Debug log

No response

Notes

No response

Start with the ember driver and UART/ASH connection logs for the Home Assistant Yellow radio at /dev/ttyAMA1, using the reported RESET_WATCHDOG pattern and the requested full debug log or Ember Helper diagnostic. Compare the behavior on firmware 7.4.4, with any 7.4.1 downgrade considered against the older issue #21462; done means stable coordinator operation without recurring resets or full Zigbee2MQTT restarts.