[core][flink][spark] Support default bucket number for postpone writes

[Zouxxyy]

Purpose

• Make postpone.default-bucket-num optional. When configured, use its value exactly for partitions receiving their first real-bucket layout and for fixed-bucket overwrite writes.
• Let Spark write directly to fixed buckets for overwrite operations and snapshots without real buckets, skipping temporary uncommitted bucket -2 files. Appends to existing real-bucket layouts remain on the staged rescale path; mixed batches stay staged and use the configured default only for new partitions.
• Centralize snapshot-bound bucket-number resolution in Core and reuse it in Spark and Flink. When no default is configured, estimate from postpone.target-row-num-per-bucket or postpone.target-size-per-bucket; execution parallelism no longer determines the logical bucket count.
• Avoid unnecessary merge-on-read metadata work: plans without selected postpone splits create an empty router without scanning bucket metadata, while other plans load bucket counts and postpone statistics only for partitions containing selected postpone splits.

Tests

• Core PostponeUtilsTest and TableScanTest: 35 passed
• Spark PostponeBucketTableTest: 30 passed
• Flink PostponeBucketTableITCase#testCompactWithTargetRowNumPerBucket: 1 passed
• mvn spotless:check -Pspark3

[JingsongLi]

+1