feat(ui): container SlotCard variant + N1 slotStatus unifier

[thinmintdev]

Summary

• N1 (status vocabulary unifier): New ui/src/dash/slot-status.js with slotPhase(slot) + slotIndicatorFromPhase(). slotIndicator() delegates container slots to the container classifier; lemond slots go through the original code unchanged — all 16 slot-indicator.spec tests stay green with no modifications.
• Container card branch: runtime==="container" slots replace the device chip + backend-mismatch block with an IMAGE-TAG chip (last path segment of image ref, full ref on hover). A "container" runtime micro-tag (N5) distinguishes cold-swap behavior. Phase logic (button states) reads container_status/container_health.
• N2 cold-swap confirm: InlineSwapPopover shows "· cold restart" badge in the header for container slots and emits an info toast before firing the swap. Lemond hot-swap unchanged.
• useSlots.ts: Slot interface gains runtime, profile, image, image_status, container_status, container_health. normalizeSlot passes them through.
• A11y: @media (prefers-reduced-motion: reduce) kills pulse animation on .dot.serving/.warming/.loading. Enable-toggle gets aria-label + role=switch/aria-checked on the visible track. N3: slot-actions gets touch-action:manipulation.
• Playwright: 13-test slot-card-container-v3.spec.ts covering all container dot states + lemond regression guard + card-level chip assertions.

Closes #657

Test plan

• slot-indicator.spec.ts — all 16 lemond tests pass unchanged
• slot-card-container-v3.spec.ts — all 13 new container tests pass
• slots-v3.spec.ts — all 6 existing tests pass
• slot-lifecycle-controls-v3.spec.ts — all 3 pass
• memory-map-v3.spec.ts — all pass
• Clean Vite build (130 modules, 9.1s, no errors)

🤖 Generated with Claude Code

[thinmintdev]

VERDICT: CHANGES (2 numbered) — not merge-ready

The architecture is sound and the container variant is approve-quality, but the shared-helper refactor breaks lemond parity in two places (AC #2: "lemond slots visually unchanged"). CI is green but does NOT exercise either regression (see note), so green is not parity-proof here.

WHAT IS RIGHT (dont reopen):

• slotIndicator() lemond path is byte-identical: the only change is a top guard (runtime==="container" || container_status!=null) delegating container slots to slotIndicatorFromPhase; lemond slots fall through to the unchanged original. Same conservative pattern in the SlotCard isContainer phase branch (lemond else = verbatim original) and stateChipClass object-path (null sentinel → original).
• Container detection via container_status != null interim fallback is sound — #656 always emits container_status even before as_dict() serialises runtime/image/profile (#658). Graceful degradation when image/profile absent ("no image" / "profile:") is correct.
• Container dot states (running+health→ready/serving, starting/pulling→warming, crashed→error, stopped/!enabled→offline), N2 cold-swap confirm toast + "· cold restart" badge, N5 runtime micro-tag, a11y (prefers-reduced-motion kills pulse, toggle aria-label, touch-action:manipulation) all correct.

CHANGES (blocking):

1. isSlotLive diverges from the old LIVE_STATES for lemond slots (memory-map attribution). slot-status.js docstring claims isLive has "same semantics as old LIVE_STATES" — but it does not. Old: liveSlots = state ∈ {ready, serving, idle, warming}. New _lemondPhase().isLive:

   • warming → was LIVE, now isLive=false (→ "starting" branch)
   • idle → was LIVE, now isLive=false (→ "idle" branch returns isLive:false)
   • disabled slot with state=ready → was LIVE (old never checked enabled), now early-returns {stopped, isLive:false}
   • reverse flip: state=offline + lemonade_state=loaded → now isLive=true (old keyed only off state)
     Net: idle/warming/disabled-resident lemond slots silently drop out of the memory map; some offline-but-loaded slots newly appear. This is a regression vs the authors own stated equivalence. FIX: restore equivalence (treat idle+warming as live, ignore enabled, key off state for lemond) — OR if the exclusion is deliberate, correct the docstring and call out the memory-attribution change explicitly.

2. stateChipClass string overload now colors lemond warming amber (was grey). The string branch added "pulling","warming" to the warn set and "crashed" to err. The sole external call site (slot-modals.jsx:589, EditSlotDrawer state ReadOnlyStrip) passes slot.state as a STRING → hits this branch. A lemond slot in state="warming" (a real lemond state) now renders a "chip warn" (amber) where the old code returned the default grey "chip". Visible lemond change. FIX: keep the string branchs lemond-relevant states classified as before (dont add warming/pulling to warn for the string path), or route that strip through the slot-object overload so only container slots get the new vocab.

CI NOTE: γ-suite + ui are green, but neither regression is covered — memory-map-v3.spec uses only ready/starting slots (no idle/warming lemond slot anywhere in mock-data), and no spec asserts the state-strip chip color for a warming lemond slot. So green CI does not verify lemond parity here. (python jobs are irrelevant — no .py changed.) Worth adding a memory-map test with an idle + a warming lemond slot to pin LIVE_STATES equivalence going forward.

Once #1 + #2 are addressed this is merge-ready — everything else is solid. Unblocks #658/#659/#660.