MINOR: Fix stale comment in partition reassignment javadoc

[stanislavkozlovski]

During the cleanup phase of an overriding reassignment, we failed to mention that we still add replicas in the adding_replicas field for the initial LeaderAndIsr request

[stanislavkozlovski]

As said in the javadoc:

* Cleanup Phase: In the cases where this reassignment has to override an ongoing reassignment.
   *   . The ongoing reassignment is in phase A
   *   . ORS denotes the original replica set, prior to the ongoing reassignment
   *   . URS denotes the unnecessary replicas, ones which are currently part of the AR of the ongoing reassignment but will not be part of the new one
   *   . OVRS denotes the overlapping replica set - replicas which are part of the AR of the ongoing reassignment and will be part of the overriding reassignment
   *       (it is essentially (RS - ORS) - URS)

We do want to keep AR=OVRS because the overriding reassignment will have it too - no need to risk shrinkage of ISR during the race between this first LAIR and the subsequent LAIR that adds the proper AR/RR fields

Here is the code which shows that we send AR=OVRS

kafka/core/src/main/scala/kafka/controller/KafkaController.scala

Line 693 in f99bb04

updateLeaderEpochAndSendRequest(topicPartition, intermediateReplicaAssignment.replicas, intermediateReplicaAssignment)

cc @viktorsomogyi @cmccabe

[viktorsomogyi]

Ok, while we're at it, shouldn't this say RR = [] at the end? (And in the line below?)

kafka/core/src/main/scala/kafka/controller/KafkaController.scala

Line 565 in f99bb04

* 1 Set RS = ORS + OVRS, AR = OVRS, RS = [] in memory

[stanislavkozlovski]

Yes, good catch. Thanks!

[cmccabe]

LGTM