fix: workflow incorrectly marked as completed while nodes are still executing

[tomerqodo]

Benchmark PR from qodo-benchmark#441

---

Note

Strengthens skip propagation semantics and adds test coverage.

• In SkipPropagator.propagate_skip_from_edge, call start_execution(node_id) before enqueue_node(node_id) when any incoming edge is taken
• In _propagate_skip_to_node, recurse via propagate_skip_from_edge before mark_edge_skipped on outgoing edges (order change)
• In skip_branch_paths, now also calls propagate_skip_from_edge after mark_edge_skipped for each unselected edge
• New unit tests for SkipPropagator covering unknown/taken/all_skipped cases, recursive propagation, and branch-path skipping

^{Written by Cursor Bugbot for commit 3a2bbf9. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Debug print statement left in production code

Low Severity

A debug print statement was left in the code that outputs "Starting execution for node: {downstream_node_id}" to stdout. This debug output will pollute logs in production and is not appropriate for production code.

 

[cursor]

Wrong order breaks recursive skip propagation logic

High Severity

In _propagate_skip_to_node, the order of propagate_skip_from_edge and mark_edge_skipped was reversed. Now propagate_skip_from_edge is called before the edge is marked as skipped. When propagate_skip_from_edge analyzes incoming edges of the downstream node, the current edge still has state UNKNOWN, causing analyze_edge_states to return has_unknown: True and the function to return early without propagating the skip. This completely breaks recursive skip propagation through the graph.