x86 confirms str.find penalty on python-build-standalone, including the newest build

Native x86-64 (Xeon 8488C, c7i.2xlarge, run on the host) reproduces the aarch64 result, so Rosetta wasn't needed and shouldn't be trusted here: under Rosetta the long-string emoji JSON result actually reversed. Against Debian 3.13.5, str.find for an absent 2-char needle was 5.4x slower on PBS 3.13.5 and 5.3x on PBS 3.13.18; find of 'TODO' in real text was 4.8x on both. It is input-specific: rfind was 0.61x (faster) and the 5-char absent needle about level, unlike aarch64 where both were 3-4x slower. pyperformance over 49 benchmarks was only 1.4% and 4.4% slower (the 3.13.18 figure also crosses patch versions), because those suites barely hit absent-needle searches. Report suite geomeans and targeted operations separately, since neither one alone describes the build.