[core] Support source table FileIO for BLOB descriptors

[LsomeYeah]

Purpose

When a source table exposes BLOB values as descriptors, a downstream write must reopen the referenced objects before writing them into the target table. Rebuilding FileIO from the target table catalog context can miss source data permissions, especially for token-based catalogs.

This PR adds the optional blob-descriptor.source-table option. When configured, Paimon loads that table from the same catalog and uses its serializable FileIO to materialize descriptor-backed BLOBs. The existing target-context behavior remains unchanged when the option is absent.

Changes

• Resolve the configured source table once and use its FileIO for non-HTTP descriptor URIs.
• Pre-initialize lazy source credentials before distributing the reader factory, avoiding per-task REST token requests.
• Resolve the reader factory outside distributed Spark tasks, including postpone compaction.
• Preserve the existing public Flink mapToInternalRow API and route the new factory path through a private helper.
• Fail fast when blob-descriptor.source-table is used by a target without a catalog loader, including REST external tables.
• Document same-catalog and branch usage, option precedence, and the unsupported REST external target case.

Compatibility and limitations

• The source table must belong to the same catalog; a branch suffix is supported.
• REST external target tables are not supported because they do not provide a catalog loader.
• blob-descriptor.source-table takes precedence over other blob-descriptor.* FileIO options.
• Without this option, the existing target CatalogContext behavior is unchanged.

Tests

• BlobDescriptorReaderFactoryTest and UriReaderFactoryTest
• Flink SQL end-to-end BLOB materialization in BlobTableITCase
• Spark V1 and V2 SQL end-to-end BLOB materialization in BlobTestBase
• Spark V2 writer-factory regression test in PaimonBatchWriteTest
• Spark postpone compaction test verifying that the source table is not resolved in tasks
• Flink null-overload source compatibility and row-wrapper compatibility tests
• Configuration documentation completeness test
• Spark 4 compilation with JDK 17
• Flink 2 compilation with JDK 11

[steFaiz]

Thanks for this PR!
Does the target table store the blobs from source table as blob format or blob descriptors?
If as blob format, I think we can use blob-descriptor.XXX to specify options for CatalogContext as in #7231
If as blob descriptors, I think blob-view is designed to solve this.

[leaves12138]

1、FileIO should be transient
2、It seems strange to hold both fileIO and readers

[leaves12138]

You can just overwrite newReader() to just put your special fileIO in

[leaves12138]

If FlinkRowWrapper with uriReaderFactory(fileIO), then it can't be successfully serialize and deserialize. FileIO is not serializable.

[LsomeYeah]

Thanks for this PR! Does the target table store the blobs from source table as blob format or blob descriptors? If as blob format, I think we can use blob-descriptor.XXX to specify options for CatalogContext as in #7231 If as blob descriptors, I think blob-view is designed to solve this.

@steFaiz
Thanks for the suggestion!
The target stores an independent managed BLOB copy: it reads the blob through the source descriptor and writes it into the target’s own BLOB storage.
blob-descriptor.* only provides static CatalogContext options, while the source table’s data token is dynamically refreshed by its FileIO and cannot be reliably materialized into an options map. blob-view is a no-copy reference with different semantics.
Hope this clarifies the motivation for source-table.

[steFaiz]

@LsomeYeah Thanks! Make sense to me!

[JingsongLi]

+1