[MINOR][CORE] Fail fast in InsertTransitions when a child has no recognizable Convention

[LuciferYang]

What changes were proposed in this pull request?

InsertTransitions.applyForNode has a branch that handles the case where a child's Convention is isNone (row type and batch type both None) by returning the child unchanged. The pre-existing comment already notes that such a plan is not executable. No production plan reaches this branch today, but the silent fall-through would defer the failure to task execution. This PR replaces the silent no-op with a GlutenException that names the offending parent and child. The same latent hole exists in Transitions.enforceReq, which currently reaches assert(!from.isNone, ...) inside Transition.Factory#findTransition and raises an AssertionError without plan identity; the fix adds a symmetric pre-check there so both entry points behave the same way.

Why are the changes needed?

Defense-in-depth: keeps a code path that is currently unreachable from silently degrading if it ever does become reachable (e.g. a future GlutenPlan that inadvertently declares both types as None). No behavior change for any currently-shipping plan.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

mvn -pl gluten-core,gluten-substrait -am -Pspark-3.5 spotless:check passes. TransitionSuite gains three tests covering (1) an unexecutable child at position 0, (2) an unexecutable child at a non-first position under a binary parent, and (3) an unexecutable root plan routed through enforceReq. Each was verified against the pre-fix code and observed to fail with the expected signal. A new NoConventionLeaf fixture in TransitionSuiteBase supplies the rowType=None, batchType=None state.

[github-actions]

Run Gluten Clickhouse CI on x86

[Copilot]

Pull request overview

This PR makes transition insertion fail fast (at planning time) when a plan (or a child plan) has an unrecognizable Convention (rowType=None and batchType=None), replacing silent no-ops / assertion failures with a GlutenException that identifies the offending plan(s).

Changes:

• In InsertTransitions.applyForNode, throw a GlutenException when a child’s derived Convention is isNone instead of returning the child unchanged.
• In Transitions.enforceReq, add a symmetric root-level pre-check to throw a GlutenException rather than hitting an internal assert(!from.isNone) in Transition.Factory#findTransition.
• Add Transition.notExecutable(...) helpers plus UT fixtures and tests to validate the new failure mode and error framing.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
gluten-substrait/src/test/scala/org/apache/gluten/extension/columnar/transition/TransitionSuiteBase.scala	Adds a NoConventionLeaf test fixture representing rowType=None, batchType=None.
gluten-substrait/src/test/scala/org/apache/gluten/extension/columnar/transition/TransitionSuite.scala	Adds tests asserting planning-time failures and message framing for child and root cases.
gluten-core/src/main/scala/org/apache/gluten/extension/columnar/transition/Transitions.scala	Replaces silent handling of from.isNone with Transition.notExecutable and adds a root pre-check in enforceReq.
gluten-core/src/main/scala/org/apache/gluten/extension/columnar/transition/Transition.scala	Introduces notExecutable exception constructors used by both entry points.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Run Gluten Clickhouse CI on x86

[zhztheplayer]

but the silent fall-through would defer the failure to task execution with an error unrelated to the real cause.

Do you mean the error is thrown from Spark executor rather than at planning time?

[LuciferYang]

Yes. If this branch were hit, no transition would be inserted for a child whose Convention is isNone, so the failure would surface in the executor at doExecute / doExecuteColumnar (or codegen) time. The branch is not currently reachable in production, so this PR is a defense-in-depth cleanup rather than a fix for an observed failure. I updated the PR body to drop the overreaching wording.