Single-run summary口径 (pass_rate / judged_pass_rate / avg_latency /
connectivity split) was inlined in run(), reachable only by driving a
whole async run, and report.py recomputed judged_pass_rate independently.
Extract build_run_summary — a pure function parallel to aggregate_runs
(cross-run) and combine_case_outcome (case-level). run() now collects
material and delegates; judged_pass_rate is stored in RunSummary so the
report reads it instead of recomputing.
9c01afa79b