Measure graph performance in verify-graph, and name port holders in run-local

[bracyw]

Changes

verify-graph gets an optional performance probe, perf.sh. It reports JS heap, allocation rate and frames, plus the browser's GPU-process memory, which is where WebGL charts differ. check.js now waits for the Quick Select options before counting them, which fixes a race that failed checks 4 and 5 under load. The skill now gives the playwright-cli install steps and notes Calypso's low data rate. When every port is taken, run-local names what holds each one.

Diagram

flowchart LR
  P["perf.sh branch url topics"] --> J["perf.js in the branch's session:<br/>heap, alloc MB/s, fps, long tasks"]
  P --> G["ps: session daemon → browser →<br/>gpu-process RSS"]

Loading

Testing

• check.sh passes 6/6 on the #872 uPlot spike, where it used to fail checks 4 and 5 with "0 Quick Select options"
• perf.sh on the graph spikes: about 34–41 MB of JS heap on all four, against a 145 MB GPU process for WebGL-Plot and 62–72 MB for the rest
• serve.sh with all 11 ports taken: lists each port's pid and working directory, exits 1 and releases the lock
• shellcheck is clean on both changed scripts

Notes

• Calypso has no rate flag (simulate --help offers only topic filters), so the skill says to treat small JS-heap gaps as noise rather than adding a stress mode.
• The port range stays 4200–4210. Stale clients from other checkouts are the user's to stop.

Checklist

• ci/check.sh . and bash tests/e2e.sh pass
• Version bumped if a .delphi/manifest.yml file changed (none changed)
• No merge conflicts
• Request reviewers & ping on Slack