Benchmarks: Publish Eigen vs OpenBLAS, oneMKL and AOCL on Zen 5 (AVX512), retire the 2009-2011 pages as history

This adds a page comparing Eigen with OpenBLAS 0.3.26, Intel oneMKL 2025.3 and AMD AOCL 5.2 on twelve BLAS and LAPACK operations, measured single-threaded in float and double on an AMD RYZEN AI MAX+ 395 w/ Radeon 8060S under WSL2 (GCC 13.3, AVX-512), and relabels the 2009-2011 pages as a historical record.

• New page (content/en/benchmarks/2026-09-zen5-blas-lapack.md): AXPY, DOT, GEMV, GEMM, TRSM, SYRK, POTRF, GETRF, GEQRF, SYEV, GESDD and GEEV, with the measurement conditions, 40 charts, a ratio summary by size band, caveats and the full per-shape tables. oneMKL appears twice, as shipped and with its CPUID vendor check bypassed. The libraries were measured on 2026-09-10 and 2026-09-11, Eigen at master 15f227178 on 2026-09-16 (harness commit 9599d90fa, four passes over the library binaries).
• Data and charts: the 155 raw result files sit beside the page, and scripts/benchmarks/plot_comparison.py draws the charts and tables from them.
• Existing pages: benchmarks/_index.md lists current and historical results separately, benchmark.md and performance_monitoring carry historical banners, and how_to_run_the_benchmark_suite describes the Google Benchmark suite under benchmarks/ in place of the removed BTL harness.

The results come from the comparison harness of eigen!2903; plan and progress are in eigen#3112. Checked with a local hugo 0.151.2 server.