feat(ui): order totals, decimals, mobile modal cards, set editing

- Dashboard/Logs: Total column with 2 decimals (order_total)
- Order detail modal: totals summary row (items total + order total)
- Order detail modal mobile: compact article cards (d-md-none)
- Mappings: openEditModal loads all CODMATs for SKU, saveMapping
  replaces entire set via delete-all + batch POST
- Add project-specific team agents: ui-templates, ui-js, ui-verify,
  backend-api
- CLAUDE.md: mandatory preview approval before implementation,
  fix-loop after verification, server must start via start.sh

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -20,8 +20,9 @@ Importa automat comenzi din GoMag in sistemul ERP ROA Oracle. Stack complet Pyth
 ## Development Commands
 
 ```bash
-# Run FastAPI server
-cd api && uvicorn app.main:app --host 0.0.0.0 --port 5003 --reload
+# Run FastAPI server — INTOTDEAUNA via start.sh (seteaza Oracle env vars)
+./start.sh
+# NU folosi uvicorn direct — lipsesc LD_LIBRARY_PATH si TNS_ADMIN pentru Oracle
 
 # Tests
 python api/test_app_basic.py       # Test A - fara Oracle
@@ -30,18 +31,21 @@ python api/test_integration.py     # Test C - cu Oracle
 
 ## UI Development Workflow: Before → Preview → After
 
-For UI/frontend changes, follow this visual verification workflow:
+**OBLIGATORIU**: Respecta ordinea exacta. NU treci la pasul urmator fara aprobare explicita.
 
 ### 1. Before Screenshots
-Capture current state with Playwright MCP at target viewports:
+Captureaza starea curenta cu Playwright MCP:
 - **Mobile:** 375x812
 - **Desktop:** 1440x900
-Save to `screenshots/before/`
+Salveaza in `screenshots/before/`
 
-### 2. Plan & Preview
-- Write implementation plan with design decisions
-- Generate preview mockups if needed → save to `screenshots/preview/`
-- Get user approval on previews before implementation
+### 2. Plan & Preview — ASTEAPTA APROBARE
+1. Citeste TOATE fisierele implicate
+2. Scrie planul de implementare cu decizii de design
+3. Genereaza mockup-uri HTML/CSS statice care arata rezultatul asteptat → salveaza in `screenshots/preview/`
+4. **Prezinta mockup-urile userului si ASTEAPTA aprobare explicita**
+5. Rafineaza planul daca userul cere modificari
+6. **NU trece la implementare pana userul nu spune explicit "ok", "aprob", "executa" sau similar**
 
 ### 3. Implementation cu TeamCreate (Agent Teams)
 
@@ -57,7 +61,12 @@ Folosim **TeamCreate** (team agents), NU superpowers subagents. Diferenta:
    - Task 1: Templates + CSS (HTML templates, style.css, cache-bust)
    - Task 2: JavaScript (shared.js, dashboard.js, logs.js, mappings.js)
    - Task 3: Verificare Playwright (depinde de Task 1 + Task 2)
-4. **Agent tool** cu `team_name` spawneaza teammates care isi iau task-uri din lista
+4. **Agent tool** cu `team_name` spawneaza teammates folosind agentii predefiniti din `.claude/agents/`:
+   - `subagent_type: ui-templates` → pentru Task 1 (templates + CSS)
+   - `subagent_type: ui-js` → pentru Task 2 (JavaScript)
+   - `subagent_type: ui-verify` → pentru Task 3 (Playwright verification)
+   - `subagent_type: backend-api` → pentru modificari backend/API (routers, services, Oracle/SQLite)
+   - `subagent_type: qa-tester` → pentru teste de integrare
 5. Teammates lucreaza in paralel, comunica intre ei, marcheaza task-uri completate
 6. Cand Task 1 + Task 2 sunt complete, teammate-ul de verificare preia Task 3
 
@@ -68,6 +77,13 @@ Folosim **TeamCreate** (team agents), NU superpowers subagents. Diferenta:
 4. Raporteaza discrepante la team lead
 5. Verifica ca desktop-ul ramane neschimbat
 
+#### Bucla de corectie (responsabilitatea team lead-ului):
+1. Dupa ce verify-agent raporteaza, **team lead-ul analizeaza discrepantele**
+2. Pentru fiecare discrepanta, creeaza un nou task de fix si spawneaza un agent sa-l rezolve
+3. Dupa fix, spawneaza din nou verify-agent pentru re-verificare
+4. **Repeta bucla** pana cand toate verificarile trec (after ≈ preview)
+5. Abia atunci declara task-ul complet
+
 ```
 screenshots/
 ├── before/     # Starea inainte de modificari