df.instance_executions() can return no rows for completed instances

[pinodeca]

Summary

df.instance_executions(instance_id, 1) can return an empty rowset for some instances that are otherwise reported as completed by df.list_instances() and aggregate metrics.

Expected behavior

Every completed instance should have at least one execution row retrievable through df.instance_executions(instance_id, 1), or the function should return an explicit error explaining why execution history is unavailable.

Actual behavior

For a subset of completed instances, df.instance_executions(instance_id, 1) returns no rows. This makes it impossible to distinguish "no execution history exists" from "execution history lookup failed".

Repro shape

1. Run enough workflows to produce multiple completed instances.
2. Enumerate completed instance IDs with df.list_instances().
3. For each completed instance, call df.instance_executions(instance_id, 1).
4. Observe that some completed instances return an empty rowset.

Impact

Per-instance latency and execution-history aggregation can undercount completed work or silently drop instances from diagnostics.