Clean up cluster tables at startup

When a Nuxeo node is shutdown abnormally (kill -9 or OOM for example), its data are not removed from the cluster_* tables and the next startup will try to add a new entry in the cluster_nodes table with the same id as an existing entry.

This may have a bad impact on the invalidation mechanism (too many entries added to cluster_invals).

NXP-23264 will ensure it is not possible to have several entries with the same id.
Some code should also do the cleanup at startup not to fail when adding an entry for the current node in the cluster_nodes table.