[Segment Replication] Differentiate post merge checkpoints

[mch2]

Is your feature request related to a problem? Please describe.
Today SegRep uses the ReplicationCheckpoint to compute staleness of replicas, report stats, and enforce backpressure for lagging replicas. It would be great to differentiate in metrics & exclude from backpressure computations the lag between checkpoints that are strictly post-merge refreshes where the searchable doc count does not change.

Describe the solution you'd like
While this data should still be surfaced it should be excluded from backpressure computations.
/_cat/segment_replication should still show ongoing lag but have a separate column to identify the lag for syncing to a merged checkpoint.
This should also be surfaced through _nodes/stats API as a separate metric for both bytes & time lag.

Perhaps we can restructure the cat SR api to show a row for each checkpoint the replica is lagging and a isMerge column.

Describe alternatives you've considered
Leaving as is / silently excluding these checkpoints.

• Investigate the underlying Lucene internals to design the new NRT checkpointing behavior  #11462
• Support copying of files that are not associated with any ReplicationCheckpoint out to replicas. #11619

[mch2]

Rather than differentiating checkpoints - we should make an implementation of IndexWriter.IndexReaderWarmer similar to Lucene NRT's PreCopyMergedSegmentWarmer.

At this point the gaps are as follows:

• Support copying of files that are not associated with any ReplicationCheckpoint out to replicas. This should be abstracted to support both node-node and remote store based copy.
• Create a new function through IndexShard to handle initiating the copy that does not ack until all replicas are current.
• Pass the function to an implementation of IndexReaderWarmer that can be wired into EngineConfig so that the warmer can be set in IndexWriterConfig.