[SPARK-56331][UI] Truncate long node labels in SQL plan visualization

[yaooqinn]

What changes were proposed in this pull request?

Truncate DOT graph node labels to 36 characters using middle-ellipsis when node names are long (e.g., scan nodes with fully-qualified catalog.schema.table paths from custom catalogs).

The full name remains available in:

• Tooltip (hover over the node)
• Side panel (click the node)

Example:

Before: ScanTransformer parquet spark_catalog.longschemaname.catalog_sales  (60+ chars)
After:  ScanTransformer pa…catalog_sales  (36 chars)

Why are the changes needed?

When custom catalogs generate long schema/table identifiers, the plan visualization graph becomes excessively wide and hard to read. Most operator labels are under 30 characters; only scan nodes with fully-qualified paths exceed 36 characters.

Does this PR introduce any user-facing change?

Yes. Long node labels in the SQL plan visualization are now truncated with a middle ellipsis. The full name is still accessible via tooltip and side panel.

How was this patch tested?

• Updated existing test for truncated label in DOT output
• Added 4 new tests:
  • Short labels unchanged (Sort, Filter, HashAggregate, null, empty)
  • Long labels truncated with middle ellipsis (boundary, realistic scan names)
  • DOT label truncated while tooltip keeps full name
  • JSON side panel details keep full untruncated name

All 6 tests pass in SparkPlanGraphSuite.

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

Generated-by: GitHub Copilot (Claude Opus 4.6)

[yaooqinn]

Screenshots

Plan visualization with truncated scan node labels:

Graph nodes show truncated labels with middle-ellipsis (...), keeping the operator prefix and table name:

• Scan parquet spa...fault.fact_table (was: Scan parquet spark_catalog.default.fact_table)
• Scan parquet spa...efault.dim_table (was: Scan parquet spark_catalog.default.dim_table)

Side panel shows full untruncated name on click:

Clicking a truncated node opens the side panel with the full name Scan parquet spark_catalog.default.fact_table and all metrics.

What's NOT truncated:

• Short operator names: Sort, Filter, HashAggregate, BroadcastHashJoin, Project, Exchange — all unchanged
• Labels without qualified paths — Execute InsertIntoHadoopFsRelationCommand stays full

Truncation only activates when:

1. Label length > 36 characters, AND
2. Label contains a dot-separated qualified name (word.word.word pattern)