[core] Support compact for chain table

[Stefanietry]

Purpose
Linked issue: #7887

Purpose

Support compact for chain table

Tests

org.apache.paimon.spark.SparkChainTableITCase#testChainMerge

[JingsongLi]

Review: [core] Support compact for chain table

Critical Bug: Resource lifecycle in mapPartitions closure

In ChainMergeProcedure.java, the BatchTableWrite and IOManager are created before the while (splitIterator.hasNext()) loop, but write.close() and ioManager.close() are called inside a finally block within the loop iteration. If a Spark partition receives more than one split, the second iteration will attempt to use already-closed resources and fail. Additionally, prepareCommit() is called per-split rather than once after all data has been written.

Fix: move the try/finally to wrap the entire while-loop, and call prepareCommit() only after all splits are processed.

Hardcoded version in pom.xml

The new dependency uses 1.5-SNAPSHOT instead of ${project.version}. This will break when the project version changes.

Typo: validataChainMerge should be validateChainMerge

Inconsistent Javadoc

The class-level doc shows CALL sys.compact(...) but the actual registered procedure name is chain_merge.

Incorrect PR title prefix

Title says [core] but all changes are in paimon-spark. Should be [spark].

Test coverage suggestion

The test should also assert that the un-merged partition in the snapshot branch remains unaffected, to guard against accidental overwrites.