[taoran92]

Purpose

Fixes #10837.

Spark queries that project or filter __paimon_partition currently fail on chain tables and query-authorized reads. Build partition metadata from visible reader rows so retained snapshot rows report the logical chain partition, while authorized reads preserve row filters, column permissions and column masks, including masks that depend on another field or return NULL.

Read and authorize the requested metadata subfields, then remove any added ordinary fields from the Spark output. Preserve nested pruning and nullable masked partition fields, and keep fields added only for metadata out of ordinary runtime-filter attributes. Document the authorization scope when nested schema pruning is disabled.

Tests

Passed locally with Spark 3.5.8 and JDK 11:

• SparkPartitionMetadataAuthTest and SparkChainTableMetadataTest: 22 cases covering ordinary, chain and fallback reads, row filters, denied columns, constant/field-dependent/NULL masks, both nested-pruning settings, metadata predicates, aggregates, limits, joins, typed partitions and blob descriptors.
• SplitUtilsQueryAuthTest: 4 cases covering visible partition values, split wrappers and reader cleanup.
• 16 additional SQL assertions covering Parquet, ORC and Avro, typed and NULL partitions, empty partition structs, mixed metadata projections, and NULL-mask counts and JSON output.