Fix Jellyfin WebSocket blocking HA startup for 300s

[erikpendragon]

Problem

When the Jellyfin sensor is configured, Home Assistant logs the following at every startup:

WARNING (MainThread) [homeassistant.core] Something is blocking Home Assistant
from wrapping up the start up phase. We're going to continue anyway. Please report
the following info... The system is waiting for tasks:
{<Task pending name='Task-...' coro=<JellyfinWebSocket._listen() running at
/config/custom_components/mediarr/server/jellyfin.py:105> ...>}

Startup stalls for the full 300s timeout before HA force-continues, delaying every add-on and integration that waits for Core to reach RUNNING. Secondary symptom: the Supervisor issues secret requests, then waits ~5 min before Core responds — add-ons like Duck DNS / Samba / Node-RED appear to "hang" at boot.

Root cause

JellyfinWebSocket.connect() spawns the listener with:

self._hass.async_create_task(self._listen())

hass.async_create_task() registers the task on hass._tasks, which Core awaits during the startup phase. Since _listen() is an infinite WebSocket read loop, it never completes, so the wait always times out.

The same pattern is used in _schedule_reconnect():

delay, lambda: self._hass.async_create_task(self.connect())

A reconnect that fires during the startup window has the same effect.

Fix

Use hass.async_create_background_task(coro, name=...), which is the HA-blessed API for fire-and-forget loops that should NOT block startup. Background tasks are tracked for clean shutdown but excluded from the startup wait set.

Reference: HA developer docs — Asynchronous programming.

min_ha_version

This PR also pins min_ha_version: "2023.5.0" — the release that introduced async_create_background_task. Without this, users on older HA would install a broken integration and hit AttributeError at connect time instead of HACS / Supervisor refusing the install with a clear message. 2023.5 is nearly three years old, so the floor is uncontroversial.

Test

On a HA 2026.4.3 instance with ~200 integrations and the Jellyfin sensor configured:

	STARTING → RUNNING
Before	~300s (timeout)
After	~15s

• No "Something is blocking" warning after the fix.
• sensor.jellyfin_mediarr still populates with library data.
• WebSocket reconnect logic still works (verified by killing the Jellyfin instance — reconnect fires as expected without re-blocking startup).

Diff size

2 files, 4 insertions, 3 deletions. Minimal, targeted change.