[server] Fix spurious ISR shrink after leader election

[fxbing]

Purpose

Linked issue: close #3698

A newly elected leader may remove healthy followers from ISR immediately after the first write because their newly created FollowerReplica state does not receive the normal lag grace period.

A repeated NotifyLeaderAndIsr request can also replace existing FollowerReplica objects and discard their fetch offsets and caught-up timestamps, causing the same spurious ISR shrink.

Brief change log

• Preserve existing FollowerReplica objects for unchanged assignments.
• Reset follower state when the leader epoch advances:
  • initialize ISR followers with the normal lag grace period for a new leader;
  • preserve follower LEO for a same-broker leader re-election.
• Keep follower state unchanged for repeated notifications with the same leader epoch.
• Add regression tests for the initial leader election, repeated notification, and same-broker re-election cases.

Tests

• AdjustIsrTest#testShrinkIsr
• AdjustIsrTest#testDoNotShrinkIsrAfterRepeatedNotifyLeaderAndIsr
• AdjustIsrTest#testPreserveFollowerStateOnLeaderReelection
• AdjustIsrTest, FollowerReplicaTest, and ReplicaTest: 25 tests passed.
• Checkstyle and Spotless checks passed.

API and Format

No public API, configuration, metric, RPC, or storage-format changes.

Documentation

No documentation changes. This is an internal replica lifecycle bug fix.

Generative AI disclosure

Yes. Generated with OpenAI Codex and reviewed by the human developer.