KAFKA-20361 : Fix LazyDownConversionRecordsTest in using zlib-ng

[muralibasani]

The test's write loop passed bytesToConvert - written as the remaining parameter to LazyDownConversionRecordsSend.writeTo(). This shrinks with each batch written, and the last batch gets truncated via Math.min(batchSize, remaining) when total V1 size exceeds V2 size — which happens with zlib-ng due to its longer match sizes compressing V1 format less efficiently.

Fix: pass bytesToConvert (constant) as remaining so no individual batch is ever truncated, regardless of the compression library used.

Reviewers: Chia-Ping Tsai chia7712@gmail.com

[muralibasani]

@chia7712 tried with this fix on the class to avoid those individual batch truncation.

[muralibasani]

@chia7712 I see build fails with spotless error, for unrelated change. Tried fixing in this pr.

Execution failed for task ':core:spotlessJavaCheck'.
--
  |   |   | > The following files had format violations:
  |   |   | src/test/java/kafka/server/SocketServerMemoryPoolTest.java

[chia7712]

@muralibasani would you mind opening a separate PR to fix it?

[chia7712]

@muralibasani mind rebasing to pull in #21893 ?

[muralibasani]

@muralibasani mind rebasing to pull in #21893 ?

@chia7712 it's rebased now.