CCR role reversal

Versions (relevant - OpenSearch/Dashboard/Server OS/Browser): 3.2

Describe the issue: CCR: persist leader->follower checkpoint metadata to avoid full re-sync on role change.

When a cluster that was previously a CCR leader becomes a follower, the replication job can restart from the scratch without awareness of which leader sequence numbers were already replicated. This can causes long catch-up times after role transitions.

Configuration:

Relevant Logs or Screenshots: