Replace hand-rolled stream handling with Node stream built-ins

[cristianrgreco]

Summary

Follows the k3s change in #1503 and replaces the remaining hand-rolled stream collection with node:stream/consumers and node:stream/promises.

• runInContainer reads the container output with text().
• putArchive no longer drains the dockerode response. dockerode 5 buffers it before resolving, so the drain was a no-op.
• streamToString is removed, as those were its only two callers.
• The HTTP wait strategy reads the last logs with text(). A log stream error is now caught by the existing catch.
• Selenium's saveRecording extracts the archive with pipeline(), so a corrupt archive or a failed source stream rejects instead of hanging.

Verification

• npm run format, npm run lint, npm run knip, npm run check-compiles: pass.
• npx vitest run on run-in-container.test.ts, http-wait-strategy.test.ts and generic-container.test.ts: 69 passed, 1 skipped.
• npx vitest run packages/modules/selenium/src/selenium-container.test.ts: 6 passed.
• A standalone script confirmed pipeline(source, tar.extract(dest)) extracts a tar-fs archive and rejects on a corrupt archive and on a source error.

Not breaking

streamToString was internal and never exported from the package index.ts. The HTTP wait strategy error message keeps its format; the logs are now trimmed as a whole instead of per chunk.