[docs] landing page still advertises the pre-8192 reference regime (N=12,000) and a --n 12000 quickstart

[luciferlive112116]

Summary

index/index.html still names N = 12,000 as CCO's reference regime, and hands
visitors a quickstart command pinned to it. Every other surface in the repo moved
to 8192, so the landing page — the first thing a new miner reads — disagrees
with the engines it is describing:

surface	says
strategy/cli.py / matmul/cli.py	--n default 8192
README.md:33	"Reference setup: 8192 × 8192 matrices"
BENCHMARKS.md:3	"Reference setup: 8192 × 8192 matrices"
README.md:37, BENCHMARKS.md:58	python -m eval --n 8192 --pairs 3
index/index.html	N = 12,000 — CCO's reference regime, tag N=12000, and python -m eval --n 12000 --pairs 3

Three stale spots:

1. KPI card — 3.5 TFLOP / N = 12,000 — CCO's reference regime. The figure is
   right for the stale N (2 × 12000³ = 3.456e12 → 3.5 TFLOP), which is exactly
   why it is wrong now: at the real default, 2 × 8192³ = 1.0995e12 → 1.1 TFLOP.
2. Matmul-arena tag — <span class="tag">N=12000</span>.
3. Quickstart — python -m eval --n 12000 --pairs 3, a copy-pasteable command
   that silently benchmarks a non-reference size.

Note the neighbouring cards are correct and use decimal N (N=1,000 → 2 GFLOP,
N=128,000 → 4.2 PFLOP, matching #102/#230's convention) — only the 12,000 row is stale.

Impact

Documentation honesty on the project's front door: a visitor's first command and the
headline "reference regime" number both contradict README/BENCHMARKS/the shipped
CLI default. Same class as the landing-page link repair in #128.

Fix

Re-derive the three claims from the shipped default: 3.5 TFLOP → 1.1 TFLOP,
N = 12,000 → N = 8,192, --n 12000 → --n 8192; plus a CPU test that pins the
page's advertised N and KPI to strategy/matmul's --n default so it cannot
drift again. index/ is not a protected path (guard regex / CODEOWNERS cover
eval/, docs/, .github/, dashboard/ only).