[PHJ2000]

Fixes #1036.

MessageBuffer.putByteBuffer can copy the wrong bytes from a sliced heap source, overwrite bytes before a sliced heap destination when reading a read-only source, or repeatedly overwrite the first destination byte when copying a read-only heap source into direct storage.

Include the source/destination array offsets and advance the direct destination address for each byte. Add a 64-case regression matrix across heap/direct storage, read-only sources, source/destination slices, and empty/partial copies. The assertions also cover untouched destination bytes and source position, limit, and mark. Ten cases fail before the fix; all pass afterward with both the default and universal buffer implementations.

Validation:

• OpenJDK 21: ./sbt -batch scalafmtCheckAll jcheckStyle and ./sbt -batch 'msgpack-core/testFull' passed (257 tests, 1 existing pending test).
• Java 8: the same 257 core tests passed with TEST_JAVA_HOME set to the Java 8 installation and the session setting set msgpackCore / javacOptions ++= Seq("-bootclasspath", "<jdk8>/jre/lib/rt.jar"). Without this setting, the build reused production classes linked against Java 9+ covariant NIO methods and failed with NoSuchMethodError on Java 8.
• Universal mode (JAVA_TOOL_OPTIONS=-Dmsgpack.universal-buffer=true): all 64 new cases passed. The full core suite had one existing DirectBufferAccessTest initialization error because MessageBuffer.unsafe is null; an isolated initialization probe fails identically against the original and patched classes.
• The sbt build did not discover the Jackson JUnit tests. Running them explicitly with JUnit Platform 1.14.4 found 77 tests, with 50 passing and 27 fixture-initialization failures. The exact same 27 failures occur with the original MessageBuffer class: the shared fixture uses JUnit 4 lifecycle annotations while the affected tests use Jupiter annotations.

No build settings or unrelated test fixtures are changed in this PR.