test(workspace): align result scenarios with the default Overview

[loopx-agent]

Superseded by the earlier #6607, which already covers the same Overview/Files read corrections together with the storage collection repair. Closing this overlapping proposal keeps one delivery route for those shared CI failures. The branch and its additional inactive-Tasks, keyboard and reduced-motion validation remain available as evidence; no merged or installed result is claimed.

The author validation record below is retained for traceability.

Selecting a Goal now opens Overview (#6564), which already displays its results. The managed-report and research browser scenarios still assumed that their first result read happened after opening Files, so the packaged checks failed with 2 !== 1 and 1 !== 0.

Qualify both entry points explicitly: Overview reads and verifies its result, Tasks performs no additional result reads, and keyboard navigation to Files performs a fresh inventory/detail read. Keep exact request counts and the existing stale, malformed, wrong-Goal, owner-only and zero-model-execution checks. The scenarios also exercise reduced motion and their existing desktop/390px layouts.

Validation on macOS / Node 24, using the same source-verified packaged frontend:

• Both original scenarios reproduce the hosted assertion failures; both patched scenarios pass, including their error/recovery cases.
• Scenario syntax, the two partition/inventory tests, bundle source verification and git diff --check pass.
• The semantic advisory reports no supported new vocabulary carriers; its syntax limitations still apply.

Only the two browser scenarios change. Hosted CI and independent exact-head review remain pending; this does not qualify the complete Python suite. The related refactoring pass reuses a local navigation locator in each scenario rather than introducing a shared helper for their different fixtures.