KAFKA-19996: Avoid premature ISR expansion under min ISR

[lh0156]

Summary

• Require a follower to satisfy the existing caught-up check before expanding the ISR when the partition is under min ISR.
• Prevent a follower that only reached a stale high watermark from repeatedly entering and leaving the ISR.
• Preserve existing leader-epoch expansion behavior when the partition is not under min ISR.
• Add regression coverage for a follower that reaches the HWM but remains behind the leader LEO.

Testing

• ./gradlew :core:test --tests kafka.cluster.PartitionTest (87 tests passed)
• ./gradlew :core:check --no-build-cache --console=plain (static checks passed; 3,383 tests ran, with unrelated timing failures in DynamicConnectionQuotaTest and SocketServerTest)
• Isolated DynamicConnectionQuotaTest rerun passed; SocketServerTest remained timing-sensitive locally.

Fixes KAFKA-19996