Retry setup on transient network errors instead of failing permanently

[jaypopat]

Problem

async_setup_entry converts meross_iot exceptions into ConfigEntryAuthFailed or ConfigEntryNotReady, but transport errors raised by aiohttp are not handled. They propagate out of async_setup_entry, so HomeAssistant marks the entry SETUP_ERROR. That state is never retried, since only ConfigEntryNotReady triggers retry with backoff, so the integration stays down until the user reloads it by hand.

A momentary network hiccup at startup therefore takes the integration offline indefinitely. The usual trigger is HomeAssistant starting before the network is ready, so the first devList call fails on DNS:

ERROR (MainThread) [homeassistant.config_entries] Error setting up entry https://iot.meross.com for meross_cloud
  File "custom_components/meross_cloud/__init__.py", line 420, in async_setup_entry
    await meross_coordinator.initial_setup()
  File "custom_components/meross_cloud/__init__.py", line 144, in initial_setup
  File "custom_components/meross_cloud/__init__.py", line 342, in get_or_test_creds
  File "meross_iot/http_api.py", line 476, in async_list_devices
  File "meross_iot/http_api.py", line 341, in _async_authenticated_post
aiohttp.client_exceptions.ClientConnectorDNSError: Cannot connect to host iotx-eu.meross.com:443 ssl:default [Could not contact DNS servers]

One attempt, zero retries. On my install it sat in SETUP_ERROR for three days after a host reboot with every Meross device unavailable, while other cloud integrations that hit the same blip at the same moment recovered on their own.

#599 reports the same failure with an identical traceback (same lines 420, 144, 342) and ClientConnectorError, which is the WAN-drop variant of the same bug.

Fix

Catch aiohttp.ClientError and asyncio.TimeoutError in initial_setup and async_setup_entry, and raise ConfigEntryNotReady so HomeAssistant retries with backoff.

aiohttp.ClientError is the common ancestor of both reported variants, ClientConnectorDNSError and ClientConnectorError, as well as ClientOSError and ServerTimeoutError.

Authentication handling is unchanged, so bad credentials still stop rather than retry forever.

Testing

Verified against the aiohttp shipped with current HomeAssistant (3.14.3) that all four variants above are caught by aiohttp.ClientError.

Confirmed the recovery path on a live 1.3.12 install: the entry had been in SETUP_ERROR since a reboot three days earlier and came back immediately on a config entry reload, which is what ConfigEntryNotReady would have done automatically.

I did not fault inject a DNS failure against the patched code, as that would have meant deliberately breaking name resolution on a production setup. Happy to add that if you would like it verified that way.