Improve remote availability checking

Currently the discovery/capabilities endpoints are fetched and then cached quite heavily, which leads to opening documents not working properly if the Collabora server is temporarily not available when the cache expires. Users need to do a full page reload and the cache needs to expire before a restarted Collabora server can be used for editing again.

In order to improve this we should work on handling the metadata fetching more gracefully.

• Put discovery/capabilities response into a common service
• Fetch them once every hour, if unavailable retry every minute
• If a page load does not have an active server (in capabilities)
  • Retry fetching on document open instead of downloading the file and show a proper error
  • This might need some internal wrapper around capabilities in the richdocuments frontend code to be able to update the state or a way to refetch capabilities and update them in the @nextcloud/capabilities npm package.

Start by tracing discovery and capabilities fetching in the richdocuments frontend, then inspect how @nextcloud/capabilities exposes and updates cached state. Compare the document-open path with the file-download fallback; done means unavailable servers are retried on the stated schedule and document opening reports a proper error or recovers after capabilities are refreshed.