graph_gate_for_publish_kind previously selected a scoped stale-view gate before checking exact canonical equality. This mislabeled exact semantic-disabled and minimal-indexing results and excluded them from Pareto analysis.
Return the canonical_graph gate when the full graph matches, and make summarize_group apply the same precedence when reprocessing retained result JSON produced by older harness versions. Preserve scoped stale-view handling for genuinely unequal graphs.
Add regression coverage in tests/test_benchmark_incremental_speed.py and tests/test_summarize_benchmark_results.py. Verified 122 benchmark/report tests, Ruff format and lint, git diff --check, and scripts/check-source-safety.sh.
Signed-off-by: Andrew Hundt <ATHundt@gmail.com>