Link and mention analysis tasks queued by a cluster member that never comes back are never executed

Precisions on the scope of the proposed change:

• Only the tasks queued during the observation.remote.reclaimMaxAge seconds (1 day by default) before the member left are taken over. Older tasks stay in the database, as today, and their number is logged. This avoids notifying very old mentions after an upgrade on a cluster whose member identifiers changed long ago.
• The change relies on XWIKI-25267 to know which members are gone for good, and through it on XWIKI-25245: without it, a member which merges back after a network issue can stay considered as departed. Such a member (like any member considered departed while it still runs, for example during a network partition longer than observation.remote.reclaimDelay) still holds its pending tasks in memory and executes them, so a task taken over by the cluster singleton node can run twice, and a mention can be notified twice. A member which restarts only loads the tasks still assigned to it, so it does not execute the tasks taken over.