feat: modernize dashboard panel UI with visual range gauges and decluttered telemetry

[NicoM701]

Summary

• Modernized the PlantRun Home Assistant panel UI with a cleaner, glassmorphic layout for both light and dark themes.
• Replaced flat text strips with modern metric cards featuring visual target range gauge bars (optimal zone markers and current value needles) for tent temperature, humidity, lighting, and power.
• Redesigned active plant cards with breeder badges, stage progression timeline bars, integrated telemetry grids (soil moisture, substrate temperature, EC), and Material Design icons (ha-icon).
• Removed redundant status badges and numerical text ranges to declutter the interface.
• Updated tests/frontend_harness.html with functional ha-icon MDI SVG rendering and multi-sensor preview state.

Test plan

• python -m unittest discover -s tests -p 'test_*.py' (160 tests passing)
• node --check custom_components/plantrun/www/plantrun-panel.js
• node --check custom_components/plantrun/www/plantrun-panel-views.js
• node --check custom_components/plantrun/www/plantrun-panel-styles.js
• Visual validation in browser harness (tests/frontend_harness.html) across dark and light themes.

Remaining risk

Visual changes rely on Home Assistant's native ha-icon element in production deployment.

[NicoM701]

Review: dashboard UI modernization

Solid direction — sensor cards + range gauges are a clear UX win over the flat tent strip. Tests/harness coverage and the stated pass list are appreciated. A few issues before this should land:

High

1. "Substrat-Temp" likely bound to tent ambient temperature (plantrun-panel-views.js _renderPlantCard): bindingForMetric(run, tent, "temperature") with label Substrat-Temp will show zelt/air temp if that’s the only temperature binding. Prefer an explicit substrate/soil-temp metric (or fall back to unbound) so the card doesn’t lie.
2. Hardcoded target bands as product truth: tent gauges use fixed ranges (e.g. 24–28 °C, 65–75 % RH, light/energy bands). Same for plant telemetry (moisture 45–65, EC 0.8–1.4). Wrong stage/crop → misleading green zone. Drive targets from stage/run config (or mark them as illustrative defaults in the UI).
3. _renderRangeBar divide-by-zero: if min === max, ((numVal - min) / (max - min)) is NaN/Infinity and the needle CSS breaks. Guard max > min (and optionally clamp targetLow/High into [min,max]).

Medium

4. Dead / incomplete status badge: _renderStatusBadge is unused; CSS has .zone-badge.bad but the helper never emits bad (only warn/good/neutral). Wire it into the cards or drop it.
5. Theme selector mismatch: production styles use .theme-light … while the prototype uses html[data-theme="light"]. Confirm the live panel actually gets .theme-light on an ancestor, or light-theme gauge tracks stay wrong.
6. Prototype vs panel drift: prototype sensor grid is repeat(5, …) vs panel auto-fit / 4 metrics; different minmaxs on plant gallery. Either sync or keep the prototype out of this PR so production doesn’t inherit stale layout assumptions.
7. "Letzter Eintrag" uses journal_entries[0]: only correct if the array is newest-first. If not, show the max occurred_at (or whatever the existing helper is).

Low / nits

8. Stage timeline assumes ≤5 steps via CSS repeat(5, 1fr) while stagePlan(run) can vary — odd layout if plan length ≠ 5.
9. Unbound tent cards still show hardcoded fallbackSub sample mins/maxes — reads like live history; prefer “Nicht zugeordnet” only.
10. Large prototype HTML churn in the same PR as production panel makes review noisy; consider splitting.

What’s good

• Escape via e() on user-facing strings looks consistent.
• Gauge math clamping to 0–100% is otherwise fine.
• Declutter (drop redundant badges) matches the PR goal.

Ask: fix (1)–(3) at least; clarify intent on substrate binding + whether targets are temporary placeholders. No approve/merge from me.

— Reviewed by Grok Bot

[NicoM701]

Spark review

Model: gpt-5.3-codex-spark · Commit: 8a0dab0780d4

Diff-only review (limited context from the provided patch only). I reviewed only the changed hunks and focused on concrete correctness regressions.

• P2: Hardcoded German stage names break plant stage timeline for non-German/migrated data — custom_components/plantrun/www/plantrun-panel-views.js:207
_renderPlantCard builds the timeline from a hardcoded German stages list and matches recorded.has(stg.toLowerCase()) / stage.toLowerCase(). If run data is in English (e.g. older installs with current_stage: "Seedling" and stage_plan: ["Germination","Seedling",…]), no segment is marked active or done, so the progress bar is always blank/future-only. This is a visible correctness regression when opening legacy or non-German datasets.

• P2: Plant card shows possibly stale journal entry as "latest" — custom_components/plantrun/www/plantrun-panel-views.js:244
_renderPlantCard uses run.journal_entries?.[0] directly for "Letzter Eintrag", assuming index 0 is the newest row. If the backend returns journal entries in insertion/chronological ascending order (common pattern), this renders the oldest entry and mislabels "Letzter Eintrag" after users log more recent entries.

Automated diff-only review. No tests executed; this is not a merge approval. Posted by the local Spark reviewer using the PR author’s GitHub account.