Improve reliability of session timeout spec

[aduth]

🛠 Summary of changes

This tries to improve the reliability of a spec which has seen intermittent failures in builds.

Example failure: https://gitlab.login.gov/lg/identity-idp/-/jobs/541333

The hypothesis is that, because the spec stubs the timeout delay to occur immediately on page load, the Warden.on_next_request call in the Ruby code happens after the request in the browser, so the forced session expiration doesn't take effect, as it would be another 30 seconds before a subsequent poll request.

The proposed solution is to increase the poll frequency so that even if the first request misses the Warden request stubbing, the next request a second later would catch it.

This also adds a couple other unrelated improvements:

• Narrow the effect of the session check delay stubbing to occur only for the single page where it's needed
• Avoid unnecessary skip_trackable assignment.
  • This was needed prior to Add Sessions API controller implementation #8089 since the session expiration was tracked with an independent session variable, not Warden's last_request_at, and the old /active endpoint actually extended the Warden session. See related discussion at Update session timeout poll response handling to omit timeout on inactive session #8521 (comment). The new endpoint assigns this property to avoid extending the Warden session.

📜 Testing Plan

Build should pass.