docs: add a clock-matched 8-core benchmark and the post-unlock core map

[Weijtmans]

Follow-up to #41, as requested in the merge comment: the taskset A/B method and the stress-ng numbers as their own PR, plus the core mapping.

What this adds to the Performance section:

• Where the new cores land. Firmware does not put them at CPU 3 and 7 to match the mask bits, it enumerates them last: the original six cores sit at CPUs 0-5 with SMT siblings 8-13, the new pair at CPUs 6-7 with siblings 14-15. This mapping is what makes a one-boot A/B possible, and nothing about the mask suggests it.
• The clock-matched benchmark. taskset -c 0-5,8-13 reproduces the stock configuration on the same boot, same silicon, same clocks. Result: 12,205 vs 16,711 bogo ops/s, +36.9%, in line with the +33.3% the core count predicts. Both runs passed --verify with zero failures, so the run doubles as a wrong-results check on the new cores.
• The 7-zip numbers stay on the page under their own heading as an indicative second data point, with the existing caveat about the overclocked baseline.

mkdocs build --strict passes.

Tested by: @Weijtmans. BC-250, Bazzite (Fedora Atomic 43), kernel 6.17.7-ba29.