[core] Add public binary serialization for BucketVectorSearchSplit

[JunRuiLee]

Purpose

BucketVectorSearchSplit is only Java-serializable today, so a reader outside the JVM cannot read it. Pushing PK-table vector search into a native engine needs a byte form: a Java planner enumerates bucket splits through PrimaryKeyVectorScan -- the ANN current-segment decision needs the bucket's complete active-file set, which an ordinary table scan split does not carry -- and dispatches each to a native worker. The reader is apache/paimon-rust#746; there is no Java production call site yet.

This follows IndexedSplit: magic and version, the nested DataSplit, then the split's own state.

i64 magic = "PKVSPLIT"; i32 version = 1
DataSplit // DataSplit#serialize
i32 payloadCount; IndexFileMeta* // IndexFileMetaSerializer#serializeList
i32 rangeFileCount; rangeFile*
rangeFile = utf fileName, i32 rangeCount, (i64 from, i64 to)*

• Payloads reuse IndexFileMetaSerializer, so IndexFileMeta.SCHEMA stays the only source of truth for what a payload carries. It has no version of its own, so a change there has to bump this one.
• Row ranges are written sorted by file name: the map's iteration order is its construction order while its equality is not, so two splits that compare equal would otherwise serialize differently.
• writeObject/readObject delegate to this codec instead of calling defaultWriteObject, leaving one serialized form instead of two. That changes the object-stream bytes, which is free while the class is unreleased (added in [core] Search primary-key vector indexes #8579, no tag contains it): both call sites, FlinkPrimaryKeyVectorRead and SparkPrimaryKeyVectorRead, produce and consume those bytes inside one job.

Read-side only; no existing write or planning path changes.

[JingsongLi]

Do we need to do so many things? Maybe just serialize rowRangesByFile and remove defaultWriteObject is OK?

[JunRuiLee]

Do we need to do so many things? Maybe just serialize rowRangesByFile and remove defaultWriteObject is OK?

Thanks for suggestion. I reworked to follow IndexedSplit (magic + version + nested DataSplit), payloads reuse IndexFileMetaSerializer, and writeObject delegates to that one codec so defaultWriteObject is gone.

[JingsongLi]

+1