Add regression coverage for None-returning orchestrations

💡 **Feature description**

Add explicit regression tests for all supported `None`-return paths in orchestration replay:

- A sub-orchestrator whose completion event has `Result=None`.
- A top-level generator orchestrator that completes without an explicit return value.

Activity completion with `Result=None` is already covered by `test_call_activity_with_none_return`. The additional tests should verify that replay completes successfully, produces the expected `None` task value where applicable, and does not call JSON deserialization with `None`.

💭 **Describe alternatives you've considered**

Relying on the shared guard in `TaskOrchestrationExecutor.parse_history_event` and the existing activity test leaves the separate `SUB_ORCHESTRATION_INSTANCE_COMPLETED` branch without direct regression coverage. A top-level no-return test would also preserve the behavior that originally made issue #260 difficult to diagnose.

**Additional context**

Issue #260 reported the original failure. The root cause was isolated in #454 and fixed by #466, released in `azure-functions-durable` 1.2.9. Current code handles both task and sub-orchestration completion events with null results, but only the activity path has a focused regression test.

Start with the existing test_call_activity_with_none_return regression test and inspect TaskOrchestrationExecutor.parse_history_event, especially the SUB_ORCHESTRATION_INSTANCE_COMPLETED path. Add focused coverage for a sub-orchestrator with Result=None and a top-level generator with no explicit return. Done means replay completes successfully, produces None where applicable, and does not attempt JSON deserialization with None.

Written by the indexing model from the issue text.