HBASE-30390 Obtain previous cell value in writer buffer instead of copying in encoding

[Huginn-kio]

HBASE-30390

Summary

Avoid copying previous cell values during data block encoding.

JIRA: https://issues.apache.org/jira/browse/HBASE-30390

Root cause

EncodingState.beforeShipped() copies the complete previous cell with
KeyValueUtil.copyToNewKeyValue(). FastDiffDeltaEncoder then uses that copied
cell for previous-value length and content comparison. This allocates and copies
value bytes even though the encoded value is already in the writer buffer.

Fix

• Keep previous key metadata stable with a key-only materialization.
• Track previous value length separately.
• Compare FAST_DIFF values against the encoded writer buffer after shipping.
• Add focused regression coverage for shipped cells, reused buffers, and failed
  encoding paths.

Testing

• JAVA_HOME=$(/usr/libexec/java_home -v 17) mvn -pl hbase-common -am -Dtest=TestFastDiffEncodingState -Dsurefire.failIfNoSpecifiedTests=false -DskipITs test
  • Passed: 6 tests, 0 failures, 0 errors, 0 skipped.
• JAVA_HOME=$(/usr/libexec/java_home -v 17) mvn -pl hbase-server -am -DskipTests -Dcheckstyle.skip=true -Drat.skip=true -Dspotbugs.skip=true compile
  • Passed: server reactor compilation completed successfully.
• Scoped Checkstyle for the two new state classes and the regression test passed.
• apache-rat:check and spotbugs:check passed for hbase-common.

The full reactor Checkstyle command is currently blocked by existing
InterfaceIsType violations in the hbase-annotations test-classification
interfaces; none of the reported lines are part of this change.

Compatibility

• No public API impact; changes are internal.
• No configuration impact.
• No persisted-format or wire-format impact; encoded output semantics are unchanged.