Port backend.ai-auto-tester e2e scenarios to webui e2e

Analyzed test scenarios in `backend.ai-auto-tester` (a separate CLI+WebUI integration tester) against the current webui Playwright e2e suite, identified coverage gaps vs. `main`, and ported the applicable scenarios in as new specs under `e2e/`.

**Boundary principle**: deterministic UI reflection belongs in webui e2e; cluster-timing/load-dependent real behavior (replica scheduling, autoscaling triggers, real inference) belongs in auto-tester. Replica scheduling latency was directly measured at ~40s to 20+ minutes on the shared QA cluster, making it unsuitable for a CI-practical bounded wait – this is why the deployment specs stop at confirming a revision is attached rather than waiting for a scheduled replica.

Sub-issues cover the individual spec files ported under this initiative.

JIRA Issue: FR-3293