chore(docs): report how long a style evaluation run takes

The runner timed nothing, so the only record of a run's duration was the
timestamp on its output file. Measure it instead.

The summary now carries elapsedSeconds, concurrency, and pagesPerSecond.
Two runs of the full corpus came in at 36.9s and about 58s, so the figure
is a range.

apps/docs/scripts/style-eval/evaluate-all.mjs

@@ -37,6 +37,8 @@ const slugs = (await readdir(GUIDES, { recursive: true }))
 
 console.log(`Evaluating ${slugs.length} guides at concurrency ${CONCURRENCY}.`)
 
+const startedAt = Date.now()
+
 const pages = []
 const failures = []
 let done = 0
@@ -110,6 +112,7 @@ const reportPages = pages
   .sort((a, b) => a.overall - b.overall)
 
 const totalCost = pages.reduce((sum, page) => sum + Number(page.cost ?? 0), 0)
+const elapsedSeconds = Math.round((Date.now() - startedAt) / 100) / 10
 
 // A human sample for judging whether the scores mean anything. Each entry links
 // to the published page so a reader can open it and disagree with the score.
@@ -133,6 +136,9 @@ const report = {
     pagesFailed: failures.length,
     inputTokens: pages.reduce((sum, page) => sum + (page.usage?.inputTokens ?? 0), 0),
     cost: `$${totalCost.toFixed(4)}`,
+    elapsedSeconds,
+    concurrency: CONCURRENCY,
+    pagesPerSecond: Math.round((pages.length / elapsedSeconds) * 10) / 10,
     byRubric,
     byCategory: Object.fromEntries(
       Object.entries(byCategory)
@@ -152,7 +158,9 @@ const destination = join(DOCS_ROOT, 'style-eval-report.json')
 await writeFile(destination, `${JSON.stringify(report, null, 2)}\n`)
 
 console.log(`\nWrote ${destination}`)
-console.log(`  ${pages.length} evaluated, ${failures.length} failed, ${report.summary.cost}`)
+console.log(
+  `  ${pages.length} evaluated, ${failures.length} failed, ${report.summary.cost}, ${elapsedSeconds}s at concurrency ${CONCURRENCY}`
+)
 for (const [name, stats] of Object.entries(byRubric)) {
   console.log(`  ${name.padEnd(20)} mean ${String(stats.mean).padStart(5)}%`)
 }