perf: use worklist algorithm for reaching definitions

[omsherikar]

Summary

compute_reaching_definitions iterates all nodes until fixed point, even when only a subset changes between iterations. This can cause unnecessary passes on large graphs.

Suggested direction

• Switch to predecessor/successor-driven worklist: initialize with all nodes, then enqueue successors only when OUT set changes.
• Preserve current transfer functions (GEN/KILL semantics) and returned structure.

Acceptance

• Same results as current implementation on existing fixtures.
• Fewer transfer evaluations on sparse-change CFGs.
• Add benchmark or instrumentation test demonstrating reduced iterations.

Code

• refactron/analysis/data_flow.py — compute_reaching_definitions