Addresses #18654.

A Playwright report and a CI run page don't currently say what they ran against. Answering "which RStudio was this?" means digging through a job log for the installer filename.

The harness now reads the build under test from the live session and reports it in two places:

• The GitHub Actions run summary, as a "Run under test" block at the top of the page.
• The Playwright HTML report, as report metadata.

Both carry RStudio's long version and release name, the R version, the edition (Desktop or Server), and the OS with its architecture:

Ubuntu 24 (x86_64) Desktop · RStudio 2026.09.0-daily+124 "Autumn Hawkbit" · R 4.6.1

The RStudio and R versions come from the window.rstudio.version automation bridge; the release name comes from .rs.api.versionInfo(), which is the only place it's exposed. A source build reports its synthetic version (99.9.9-pr+<run> in CI), so it can't be mistaken for a daily.

In the report, each engine gets its own metadata entry keyed by OS and edition, so a merged multi-engine report keeps every engine's line instead of the last one overwriting the rest. The OS name comes from a new PW_OS_LABEL, set per engine; on a Linux engine it reuses the existing pretty_name from the engine config, so all sixteen are covered by one line.

Two supporting changes:

• os-e2e-deps no longer writes its own ### R version summary block. It runs before the tests, so leaving it would duplicate the R version and push the new block off the top of an append-only summary. Its r-version.txt artifact and step outputs are unchanged, so the certification workflow still reads what it expects.
• The S3 report URL published by os-publish-e2e-report now ends in #?show-metadata-other. Playwright renders metadata keys it doesn't recognize only behind that flag.

Verified on Ubuntu 24 x86_64: run 33005666605.