landing: refresh the performance claims

[tannevaled]

The #perf paragraph and the trust bar still carried claims from before v0.6.0. Text only, no brand or CSS change.

• vs 16-thread OpenBLAS: "MatMul 0.55× at 1024², Dot/mat·vec 0.16×" becomes the current same-run figures: MatMul 0.42× at 512² and 0.51× at 1024², but 1.44× faster at 256²; Dot 0.69×; mat·vec 0.52×. Source: go-ndarray/ndarray docs/perf.md.
• Other targets: adds the RISC-V X60 (RVV) and IBM z17 (s390x) results against the scalar code: Sum up to 5.8×, MatMul −31…−45% on the X60; MatMul 1024² 340 → 70 ms and Max −92…−96% on the z17.
• Trust bar: "5 SIMD targets" becomes 6 (amd64, arm64, ppc64le, loong64, riscv64, s390x).

🤖 Generated with Claude Code