[spark] Reject recursive Paimon view definitions

[Zouxxyy]

Purpose

CREATE OR REPLACE VIEW v AS SELECT * FROM v can analyze the old definition and then persist a recursive Paimon view. Reading the stored definition subsequently recurses until StackOverflowError.

Validate the original SQL before replacing metadata, preserving the existing definition when a cycle is found. Also reject legacy persisted cycles during resolution and report their dependency path. Paimon expands views without retaining Spark's logical View nodes, so the checker follows stored SQL dependencies, including nested subqueries and CTEs, and respects catalog identifier case sensitivity.

The guard points follow Spark's ViewHelper and ViewResolution. A separate internal checker enforces spark.sql.view.maxNestedViewDepth and reuses validated subtree depths only within one resolver traversal. This avoids repeatedly scanning a nested view chain while still checking shared dependencies reached through deeper paths.

Tests

• PaimonViewTest: 20 tests passed on each of Spark 3.5 / Java 8, Spark 4.0 / Java 17, and Spark 4.1 / Java 17.
• Covers direct and indirect cycles, legacy persisted cycles and recovery, EXISTS/IN/scalar subqueries, CTE dependencies, mixed-case identifiers, valid CTE shadowing, and shared dependencies.
• Regression coverage bounds catalog lookups for a nested view chain, detects metadata changes on the next resolver invocation, and checks nesting limits through shared dependencies.
• Final Spark 3 and Spark 4 Maven test builds passed with Checkstyle and Spotless enabled.
• For the local Spark 4.1 run, only the generated test Hive configuration used port 19092 because port 9092 was occupied.