feat: publish native iOS and macOS benchmark pilots

[janicduplessis]

Description

The website shows React Native results but omits the completed standalone native pilots. Add the accepted Luna / Stim 1.20.2 comparisons for native iOS and macOS. Fixes #3684.

Solution

A separate section reports complete task time: iOS 170 s with Stim versus 307 s control, and macOS 71 s versus 145 s. Each is explicitly one pilot pair, including retries and proof capture, with expandable environment details and verified screenshots. These results do not enter the React Native Settings-ready charts or claim a general compiler speedup.

The exporter rejects unaccepted, mismatched, or modified evidence and incomplete cleanup. It emits only public metadata and selected proof images. Native Android is supported by the exporter and view but stays unpublished until an accepted pair exists.

Test plan

• Exported the retained iOS/macOS campaign pairs and visually checked all four proof images; Android has no published pair.
• Regression tests cover publication gates, timing consistency, and Android eligibility.
• Full checks pass: formatting, lint, build, typecheck, knip, 8,917 unit tests (20 existing skips), runtime checks, 115 end-to-end tests, and website production build.
• Visually checked the rendered cards and expanded iOS proof. Website TypeScript reports the same 31 pre-existing diagnostics as the base checkout.

[janicduplessis]

Fresh independent review is clear at f90d832: issue, exporter acceptance/hash checks, provenance, tests, guidance, and presentation reviewed; 10 exporter tests and 1 presentation test passed independently. Desktop and 390px mobile rendering were visually checked, including expanded proof. Native Android remains withheld until an accepted pair is available.

[janicduplessis]

Windows CI hit an unchanged PowerShell detached-entry startup timeout (tracked in #3688). Retrying the failed job on the same reviewed head; no implementation changed. The Node 22/24 checks passed.