Fix id collisions between AssignUniqueId operators

[hqbhoho]

Description

AssignUniqueId embeds only the stage id and partition id of the task in the id mask, while row ids were drawn from a per-factory pool starting at zero. Two AssignUniqueId nodes executing in the same task therefore assign the same ids to different rows.

MERGE hits this when the join is colocated with the partitioning of the target table (e.g. a bucket-partitioned Iceberg target with join_distribution_type=PARTITIONED): the target scan stays in the join stage, so the target and source AssignUniqueId nodes run in the same task. After the right join coalesces the two id columns, MarkDistinct over (unique_id, case_number) treats unrelated rows as duplicate matches and fails with a false MERGE_TARGET_ROW_MULTIPLE_MATCHES.

The fix shares a single row id pool (AtomicLong) across all AssignUniqueId operator factories in a task, held by LocalExecutionPlanContext and passed to every factory. Ids are then unique within the whole query: the stage/partition mask separates tasks, and the shared pool separates operators within a task. This follows the operator-level direction suggested by @Praveen2112 in the issue, and subsumes the factory-duplicate() pool sharing from commit 0e880c8.

There is no execution-time cost: the per-row path is unchanged, and the shared pool is touched once per 2^20 ids per driver (a single getAndAdd).

Validation:

• New TestAssignUniqueIdOperator covers id disjointness for factories sharing a pool and for duplicated factories.
• New TestIssue30639 reproduces the issue end to end; it fails on current master with One MERGE target table row matched more than one source row and passes with this change. The test pins bucket_execution_enabled and asserts both AssignUniqueId nodes plan into one fragment, so it cannot become vacuous if planner defaults change.
• TestMerge passes unchanged; the plan shape is not affected.

Additional context and related issues

• Fixes MERGE fails on a Partitioned Iceberg tables without duplicates with MERGE_TARGET_ROW_MULTIPLE_MATCHES #30639
• Supersedes Fix MERGE row ID collisions between target and source #30937, which addressed only the MERGE consumer by widening the MarkDistinct key; the regression test here is based on the reproducer by @Harmuth94 from that PR.

Release notes

( ) This is not user-visible or is docs only, and no release notes are required.
( ) Release notes are required. Please propose a release note for me.
(x) Release notes are required, with the following suggested text:

## General
* Fix false `MERGE_TARGET_ROW_MULTIPLE_MATCHES` failures for `MERGE` when the join is colocated with the partitioning of the target table. ({issue}`30639`)