routercap: capacity benchmark harness for atenet-router

Open-loop ladder sweep of one router pod per Envoy CPU size: client
latency timed from scheduled send (no coordinated omission), per-hop span
breakdown, per-thread CPU, and guards that mark an arm rig-limited the
moment the harness rather than the router runs out. One command per sweep,
self-contained report.html per run; methodology in README.md.

Author: chuangw6

benchmarking/routercap/.gitignore

@@ -0,0 +1,13 @@
+# A run directory ships exactly one file: report.html, which embeds every
+# chart inline. The raw data behind it (samples.jsonl, run.json,
+# worker-cpu.jsonl per arm) and the regenerable summary.json/SVGs stay local
+# to the machine that ran the sweep.
+runs/**
+!runs/*/
+!runs/*/report.html
+
+# profile.sh needs privileged node debug access that the shipped harness
+# should not imply is routine.
+profile.sh
+
+__pycache__/