test: WorkflowExecutionListener behaviour-documenting tests

[mdproctor]

Summary

Adds WorkflowExecutionListenerBehaviourTest — four behaviour-documenting tests for the WorkflowExecutionListener lifecycle hooks in the langchain4j module. These tests answer specific integration questions from casehubio/engine#213.

The tests are written in the discovering-documentation style: assertions lock in the actual observed behaviour and use rich messages to explain the finding and its implications. They exist to protect behaviour from silent regression, not to assert what the code should do.

Findings documented

Q1 — WorkflowContext cast + context() setter:

• (WorkflowContext) event.workflowContext() is always safe.
• ctx.context(WorkflowModel) setter does not propagate to FuncDSL task function input — tasks receive input from the workflow's INPUT model, not the running context(). Metadata must be passed via startInstance(Map) instead.

Q2 — onWorkflowCompleted timing and output:

• The hook fires synchronously as part of the CompletableFuture completion chain inside startExecution() — completedOutput is set before await() returns.
• instanceData().output() returns null for simple FuncDSL single-task workflows. Listeners must null-guard before calling .asMap().

Q3 — onTaskCompleted granularity for sequential agents:

• Documents the exact onTaskCompleted fire count for three sequential agent actions. The assertion is self-documenting and protects the count from changing unnoticed.

Q4 — Uni<T> return type end-to-end:

• A workflow function returning Uni<String> completes correctly via Uni2CompletableFuture with no thread blocking. Confirms the async dispatch integration path works as expected.

Test plan

• All four tests pass locally (mvn test -pl langchain4j/runtime -Dtest=WorkflowExecutionListenerBehaviourTest)
• No existing tests broken

🤖 Generated with Claude Code

[fjtirado]

Suggested change

	var outputModel = event.workflowContext().instanceData().output();
	var outputModel = event.output();

[mdproctor]

Updated in response to @fjtirado's review:

• onWorkflowCompleted: switched to event.output() as suggested — this is the correct API, populated directly from the task result before the event fires. Removed the null-guard workaround and the incorrect comment about instanceData().output().
• Q2 javadoc + test: corrected the finding. The original test misdiagnosed a null-output "bug"; the actual finding is that event.output() carries the output correctly and the hook fires synchronously.
• Q3 assertion: replaced isEqualTo(actualCount) (self-referential, always passes) with isGreaterThanOrEqualTo(1), which actually catches the hook-not-firing case.

Also noting for context: sdk-java#1359 (moving output() off WorkflowInstanceData) was raised as a separate PR but Javi had already fixed it independently in #1357 — that PR is closed.