Parity testing

Behavioral parity is defined across three layers: Excel (reference), FormulaEvaluator, and standalone exported code. Shared semantics live in excel_grapher/exporter/export_runtime/ so the evaluator and generated code stay aligned (evaluator ↔︎ export checks use tests/integration/utils/parity_harness.py). Evaluator ↔︎ Excel checks use values saved in the workbook and, when automation is available, live recalculation via xlwings that require Excel should run when automation works and pytest.skip otherwise (see tests/integration/evaluator/test_golden_master.py).