Cluster paused detection blocks garbage collection of IPAddressClaims on cluster deletion

[schrej]

When deleting a cluster, it can happen that the Cluster resource gets deleted before all IPAddressClaims are deleted. When that happens, the paused detection is unable to find the Cluster and therefore skips reconciliation, preventing the claim from getting deleted.

I'm not sure whether it makes sense to skip reconciliation when the cluster is not found. In general, there shouldn't be a situation where the Cluster resource does not exist when a claim is created. The only case would be clusterctl move, and we need to validate whether the Cluster resource is moved first when performing a move.

One solution would be to ignore not found errors when a claim is deleted. As mentioned above, we need to make sure this doesn't break cluster moves.
Alternatively we could set owner references with prevent owner deletion to attempt to keep the Machine and subsequently the Cluster resource until claims are cleaned up.

We should update the CAPI book with the solution we choose.

/kind bug