[SPARK-56121][SQL] Use dedicated EmptyRelation node for static PropagateEmptyRelation

[liuzqt]

What changes were proposed in this pull request?

In previous PR #46830, I introduce dedicated node for empty relation

• EmptyRelation logical node
• EmptyRelationExec physical node
  and wrap the eliminated plan segment in AQEPropagateEmptyRelation

In this PR, I extend this change to static PropagateEmptyRelation rule.

Note: this doesn't 100% remove the usage of empty LocalRelation for empty, there're still some other rules which can produce empty LocalRelation, and we still recognize the empty LocalRelation as an empty relation.

Why are the changes needed?

Better observability, in both spark UI, and explain plan string, we're able to see the eliminated plan segment.

Plan string

• before

== Physical Plan ==
LocalTableScan (1)


(1) LocalTableScan
Output [2]: [k#1L, v#7]
Arguments: <empty>, [k#1L, v#7]

• after

== Physical Plan ==
EmptyRelation (1)
+- Project [k#1L, 1 AS v#7]
   +- EmptyRelation
      +- Join Inner, (k#1L = k#6L)
         :- RepartitionByExpression [k#1L], 8
         :  +- Project [id#0L AS k#1L]
         :     +- Range (0, 50000, step=1, splits=Some(32))
         +- LocalRelation <empty>, [k#6L]


(1) EmptyRelation
Output [2]: [k#1L, v#7]
Arguments: [plan_id=2]

Spark UI

• before

• after

Does this PR introduce any user-facing change?

Yes

• spark ui change
• explain string change

How was this patch tested?

• updated exising UTs
• verified Spark UI change

Was this patch authored or co-authored using generative AI tooling?

Yes