docs: align styler page with measured 2.6.x numbers (drop stale 10.3 KB / marketing claims)

[vitbokisch]

Why

The styler docs had several stale or marketing-shaped claims that didn't match measured reality:

• Bundle size was reported as ~10.3 KB gzipped (and ~10.34 KB in another spot) across 4+ pages. A fresh build of the current 2.6.x source is 4.82 KB gzipped / 13.65 KB minified / 33.97 KB unminified.
• The "45% smaller than sc / 40% smaller than emotion" headline was computed against that stale size.
• The micro-benchmark table claiming "2.8–1034x faster" came from an older vitest-bench iteration that's no longer the canonical suite.

What this PR does

• Bundle: replaces the stale claim everywhere it appears (content/docs/index.mdx, connectors/index.mdx, core/index.mdx, styler/api.mdx, styler/index.mdx) with the fresh 4.82 KB gzipped number.

• Bundle table: keeps the competitor comparison but as raw KB only (no "X% smaller" framing).

• Competitive bench: replaces the old micro-benchmark table with the actual css-perf.bench.tsx numbers — 3-run medians on bun 1.3.13 / react 19.2.6, equivalent work (sc wrapped in ServerStyleSheet):

  Scenario	styler ops/s	emotion	sc 6
  SSR static (500 renders)	719	335	210
  SSR dynamic (500 renders)	434	307	202
  SSR themed (500 renders)	363	286	184
  CSR mount (100 mounts)	13912	13817	13702
  CSR update (100 updates)	13108	13032	13239
  CSR many (50 distinct comps)	18933	18302	5705

  Plus a "vs best competitor" column so readers can see the gaps at a glance.

• Helper-level microbench: adds a short section with the measured helper-vs-helper deltas from perf-audit-bench.tsx (_staticResolved +149%, hashUpdate +46% long / +15% short, HTML_PROPS in obj +19%) with the honest caveat that they don't move the headline bench because the headline bench doesn't exercise those paths.

• CI-gated framing: notes that the bench workflow fails on >10% regression vs main on the same runner (which caught and forced revert of one PR #253 candidate).

• Drops "marketing-shaped" superlatives ("blazing fast", "high-performance" descriptor in the table) in favour of measured statements.

Test plan

• bun run build — Next.js docs site builds (67 static pages) without errors
• No lint changes
• No broken links (only existing in-repo links updated)

Companion to the perf work shipped in ui-system #233, #242, #244-251, #253, #254.

🤖 Generated with Claude Code