Added watch folder again

[qstokkink]

Fixes #8280

This PR:

• Adds the watch folder back again.
• Fixes a bug in the settings endpoint that would not allow new settings to be written.
• Removes the unused watch_folder_corrupt_file notification.
• Removes the unused family filter and hide tags settings.

[egbertbouman]

I'm not sure I understand the new logic. Why are we removing downloads now?

Also, why use replace_task? What's wrong with processing torrents one at a time?

[qstokkink]

Why are we removing downloads now?

🤔 In retrospect, maybe this was overly ambitious. I turned this into folder synchronization, while all we needed was a watch folder. I'll remove this.

Also, why use replace_task?

If the previous task is still pending and we re-enter the loop (i.e., for some reason this has not completed after 10 seconds), as opposed to catching a RuntimeError if the task already exists, I chose to cancel the previous task and retry with a new task. I opted for this because there is a (extremely small) chance that a new .torrent with newer metainfo was added for the same infohash, and try .. except RuntimeError .. is less aesthetically pleasing to me.

[qstokkink]

@egbertbouman fixed up.

If you don't accept my reasoning and/or feel strongly about avoiding replace_task, please offer an alternative implementation.

[egbertbouman]

The interval_runner of register_task waits 10s seconds in between tasks, so if you simply await it (and don't create an additional task), there is no problem. Having said that, I don't have strong feelings about this. I was just curious.

[qstokkink]

I don't really see how you could avoid creating additional tasks, with start_download and TorrentDef.load being async. That said, if you do see a way to make this sexier/learner/meaner, that would be cool. For now, I'll merge this - but feel free to make a follow-up PR with sexification.

[egbertbouman]

I don't really see how you could avoid creating additional tasks, with start_download and TorrentDef.load being async.

You could just make check async, which shouldn't be a problem since it's being run by the TaskManager.

[qstokkink]

Ah! That's the missing link. I never once thought about making check async.