Refresh the large-bibliography benchmark results (Phase A4)

[dknauss]

Summary

Phase A, ticket 4 of docs/planning/refactor-and-standards-memo.md.

This reruns the May 2026 benchmark against 1.11.0 plus main, which now has the rewritten CSL styles and the particle-aware sort. It uses the existing harnesses: scripts/benchmark-formatter.php for PHP and npm run benchmark:perf for JS, 5 runs per cell. The results doc gets new tables, an honest machine note, and a comparison section.

Worst p95 at 200 entries:

Layer	Worst p95	Budget
PHP formatter (citeproc-php)	45.3 ms (author-date, cold)	250 ms
JS style switch	14.8 ms	250 ms
JS mutation	26.8 ms	150 ms

Every cell is at least 5× under budget.

Against May. Absolute times are about 2–3× higher in PHP and about 1.3–1.8× higher in JS. The runs came from different machines (a Linux cloud container on a 2.1 GHz Xeon, not an M-series Mac), so the doc treats the change as direction only, not as a regression.

The one real change: author-date is now the slowest style at 200 entries (about 45 ms against notes' 36 ms), where in May they matched. That could be the rewritten style or noise. Only a same-machine run of old styles against new ones would settle it.

A follow-up this doesn't cover. This PR's sibling, #152, ran the runtime-matrix timing of a 200-entry /format request over real HTTP (nginx, PHP 8.3). It measured a p50 of about 205 ms for Chicago notes, against about 71–73 ms for author-date and IEEE. In-process, the same style formats in about 34 ms. So most of the HTTP cost for notes sits outside citeproc's render: request boot, sanitization, cache miss, or CiteProc setup for that style. It's still under the 250 ms budget, but it's the closest number to it anywhere. Worth profiling separately; it's not something this doc PR should chase.

Validation

• Docs only. Every number comes from runs made for this PR. The raw reports went to the gitignored output/benchmarks/.
• The PHP harness already existed on main; no new code.

Checklist

• No code changed
• Numbers reproducible with the commands named in the doc

🤖 Generated with Claude Code

https://claude.ai/code/session_01KbVuMPN7YTcV27vwsG3LZ2

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.28%. Comparing base (4d638a2) to head (2757f98).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##               main     #153   +/-   ##
=========================================
  Coverage     88.28%   88.28%           
  Complexity       22       22           
=========================================
  Files            61       61           
  Lines          7290     7290           
  Branches        690      690           
=========================================
  Hits           6436     6436           
  Misses          291      291           
  Partials        563      563           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.