Prevent invalid renewals of PRRLs

Today when a PRRL is created during peer recovery it retains history from the
sequence number provided by the recovering peer. However this sequence number
may be greater than the primary's knowledge of that peer's persisted global
checkpoint. Subsequent renewals of this lease will attempt to set the retained
sequence number back to the primary's knowledge of that peer's persisted global
checkpoint tripping an assertion that retention leases must only advance.  This
commit accounts for this.

Caught by [a failure of `RecoveryWhileUnderLoadIT.testRecoverWhileRelocating`](https://scans.gradle.com/s/wxccfrtfgjj3g/console-log?task=:server:integTest#L14)

Relates elastic#41536

Author: DaveCTurner

server/src/main/java/org/elasticsearch/index/seqno/ReplicationTracker.java

@@ -499,7 +499,7 @@ public synchronized void renewPeerRecoveryRetentionLeases() {
                     if (retentionLease != null) {
                         final CheckpointState checkpointState = checkpoints.get(shardRouting.allocationId().getId());
                         renewRetentionLease(getPeerRecoveryRetentionLeaseId(shardRouting),
-                            Math.max(0L, checkpointState.globalCheckpoint + 1L),
+                            Math.max(retentionLease.retainingSequenceNumber(), checkpointState.globalCheckpoint + 1L),
                             PEER_RECOVERY_RETENTION_LEASE_SOURCE);
                     }
                 }