[com6056]

What does this do

@DonMusto found this on #175. Once the last viewer leaves, the camera keeps pushing to HA's local receiver until something stops it. He saw it run for 12 hours until the camera dropped offline, and I reproduced it here (details below). Unlike the cloud push, which lapses about 20 minutes after the last PUT_STREAMING, the local push holds one of the camera's app slots indefinitely.

• When the receiver has had no reader for 90 seconds, the camera entity sends STOPPED naming the push's URL, the same per-URL stop as fix: stop our own stream push by URL so local streaming comes back after a fallback #198. It only does that while the camera is publishing, since a stop for a push the camera isn't making is answered about 70 seconds late and could end the next one. 90 seconds is longer than the 60 second stills retry, so a still taken after a failed one finds the push running.
• A viewer arriving cancels the timer. After a stop it checks again, twice as long apart each time up to 30 minutes, since a stop can fail or be answered and ignored.
• A push that outlived an HA restart (HA doesn't remove entities on shutdown) has no recorded URL, so the receiver's own URL is used, at most twice until a URL is recorded again.
• A successful idle stop counts for the settle retry from fix: stop our own stream push by URL so local streaming comes back after a fallback #198, so reopening the camera while it's winding down gets one more try instead of a cloud fallback. Removal waits for an idle stop already in flight instead of sending a second one, and cancels the keepalive and refresh tasks before that wait so they can't act on the removed entity.

There's one more fix in here. Renaming the camera entity broke local streaming until a reload. HA removes the entity and adds the same object back, removal detached it from the receiver, and nothing attached it again. async_added_to_hass now does, and reschedules a running cooldown's end.

Changelog

• The camera's local push to Home Assistant now stops about two minutes after the last viewer leaves, instead of running until the integration is reloaded and holding one of the camera's app slots.
• Renaming the camera entity no longer breaks local streaming until the integration is reloaded.

How I tested this

All 944 unit tests and 416 aionanit tests pass, and ruff, ruff format and mypy are clean. New tests cover the stop after the last viewer and a viewer coming back, no stop while the camera isn't publishing, retries and backoff, the derived URL and its cap (including timeouts and failed sends), the settle retry after an idle stop, removal during a stop (including a keepalive in flight), a rename (including one during a cooldown), and sleep.

On my prod HA (v1.14.0-beta.4) I opened the card for 30 seconds and closed it, and 12 minutes later the camera was still pushing to the receiver at about 770 kbit/s with no reader. Then I ran this branch in a dev HA against the same camera. I opened the card (local), closed it, and about two minutes later (30 seconds for HA's own HLS output to let go, then the 90 second timer) the integration sent the stop and the push was gone within 15 seconds. Reopening the card went local on the first try, with the picture up in about 5 seconds.

Checklist

• just check passes (lint, types, tests)
• Tested with dev HA instance (just dev) and verified the change works

[com6056]

Pushed one small follow-up while re-reviewing #181. Removal now cancels the keepalive, refresh, still and cloud stop tasks before it waits for an idle stop in flight, instead of after. During that wait a keepalive could re-record the push and a refresh could re-arm the keepalive timer on the removed entity. It also puts the loop where #181's background start task needs it, since a late ACK there would record a push on a removed entity the same way. There's a new test for it, and I updated the description to match.