perf(benchmarks): slim card projection for /benchmarks hub

[Flotapponnier]

Root cause

/benchmarks passes the full Benchmark[] (37 entries) into <BenchmarkGrid />, a client component. React serialises the whole prop tree into the RSC payload baked into the HTML, including fields the card never reads: extras.series7d, extras.series30d, extras.seriesByRegion24h/7d/30d, metricPanels, methodology, abstract, faq, findings, perChainExplainer, seoIntro, bestPerChain, worstPerChain, cellRanks, providersPerChain.

Result on prod: 3.2 MB of HTML per hub render. The "8 s" reported in the brief was full transfer time over the wire, not TTFB (server side was ~0.46 s STALE / ~1.13 s cache miss).

Fix

Add BenchmarkCardData (slim) + toBenchmarkCardData projection in src/data/benchmarks.ts. Hub and category pages map to the slim shape before passing to the grid. Card/grid/mini-chart prop types updated; MiniChart accepts a structural subset so the home table can keep passing full Benchmark.

No loader change, no query change. Pure wire-payload trim.

Before / after

Local prod build, curl /benchmarks:

	Size	TTFB
Prod (main)	3.2 MB	0.46 s STALE / 1.13 s cache miss / 8.7 s total transfer
This PR (local)	283 KB	0.076 s first hit / 0.005 s subsequent

92% HTML payload reduction. TTFB stays sub-second.

Alternatives considered

• Bump revalidate 60 to 300 s: shifts the problem to fewer users, doesn't fix it. Skipped.
• Build-time prerender (force static): the page already prerenders; the cost is downstream of that.
• Refactor tryLoadLive to parallel per provider: not the bottleneck for the hub, but worth doing for individual bench pages. Out of scope.

Diff stats

6 files changed, 81 insertions(+), 10 deletions(-)

Test plan

• pnpm typecheck clean
• pnpm test 114 pass
• pnpm build succeeds, /benchmarks static HTML drops 3.2 MB to 283 KB
• pnpm lint no new warnings
• Verify staging Preview URL renders identically