* test(mobile): single-source map for per-view mobile treatment
Declare how every top-level view behaves on a phone in one place:
MOBILE_TREATMENTS is a Record<ActiveView, ...>, so adding a new view without
classifying it (bespoke / responsive / desktop-only / detail) fails the type
check. BESPOKE_MOBILE_VIEWS is derived from it instead of hand-maintained, and
a unit test keeps the two in lockstep and pins the current bespoke set so a
change is deliberate. EditorLayout consumes the derived set; behavior is
unchanged.
* ci(visual): run the desktop-unchanged gate in its own job
Split the visual-regression spec into a dedicated Playwright "visual" project,
excluded from the default chromium project the functional E2E job runs, so a
missing or platform-mismatched baseline can no longer fail every PR.
Add a Visual Regression workflow: a compare job gates PRs into main against
committed baselines (and skips with a warning until they are seeded), and a
manual seed job regenerates the baselines on the Linux runner and commits them
to a feature branch (refusing main). Baselines are platform-specific, so they
must be produced on the runner rather than locally.
Drop the stack-detail view from the gate: a fresh CI app has no stack to open
and its live log stream is not deterministic; the shell plus the four content
views still catch a desktop base-class regression.
* feat(ui): make Fleet, Settings, and the dashboard table usable on mobile
Tier 2 of the mobile pass, all gated below the md breakpoint so desktop
renders identically:
- Fleet: the tab strip scrolls horizontally and the action row (Check
Updates / Refresh / Add Node) wraps instead of clipping.
- Settings: below md the nav rail is a full-screen list; choosing a section
pushes it full-screen with a back affordance (master/detail), matching the
stack flow. Desktop keeps the two-pane layout.
- Dashboard: the fixed stack-health table scrolls horizontally on a phone.
Adds a Playwright desktop visual-regression spec (1280/1440/1920) as a
zero-desktop-change gate; its environment-specific snapshots are gitignored.
* test(ui): harden the desktop visual-regression gate
Mask only the live sidebar ticker and notification count instead of the whole
sidebar / top-bar shell, so the gate now proves that shell unchanged too.
Lower the pixel budget to 1200 (the only residual churn is in-content live
stats); run against seeded / frozen data in CI for a zero-tolerance gate. Adds
a data-testid to the activity ticker so it can be masked precisely.
Anso