fix(loadtest): measure parent-side wall in the multiprocess sweep

[JumpTechCode]

Summary

The multi-process fan-out (loadtest.multiprocess.multiprocess_sweep) gathered each slice's CommandSamples but called summarize("guarded-mp", samples, wall_seconds=0.0), so the reported throughput was always 0.0 (count/wall collapses to 0.0 when wall is non-positive). Correctness signals — oversell, retries, latencies, and the oracle verdict — were already meaningful; only throughput was degenerate.

This threads a real aggregate wall through the sweep: it times the parent-side wall around the ProcessPoolExecutor run (from dispatch until the last slice returns) and passes it to summarize, mirroring how the single-process drive() reports its wall.

Why parent-side wall

The parent-side span is the correct denominator for an aggregate throughput across overlapping workers — summing per-slice walls would double-count concurrent time. The comparative table's published throughput/p99 numbers still come from the single-process run; this is the multi-process path's own end-to-end figure.

Test

TDD: extended test_multiprocess_guarded_stays_clean with assert report.metrics.throughput > 0.0. Confirmed it fails first (assert 0.0 > 0.0, with count=48 and valid latencies), then passes after the fix. Full make verify is green (591 passed, 99.14% coverage; ruff, mypy --strict, import-linter, pip-audit clean).

Closes #96.

🤖 Generated with Claude Code