docs: update Architecture + SDK Flow with Control Pipeline + Dependency Engine

Architecture (architecture-data.ts):
- Replace document-crawler with control-pipeline (Port 8098)
- Add 9 DB tables, 5 RAG collections, 10 API endpoints
- Add edges: control-pipeline → PostgreSQL, Qdrant, Ollama

SDK Flow (steps-betrieb.ts):
- Add 4 new steps (seq 5200-5500):
  - Canonical Control Library (7-stage generation pipeline)
  - Pass 0a: Obligation Extraction (181k obligations)
  - Pass 0b: Atomic Composition (MCP-taugliche controls)
  - Dependency Engine + Evaluation (5 types, auto-generation)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

admin-compliance/app/sdk/architecture/architecture-data.ts

@@ -228,24 +228,39 @@ export const ARCH_SERVICES: ArchService[] = [
     dependsOn: ['qdrant', 'ollama', 'postgresql'],
   },
   {
-    id: 'document-crawler',
-    name: 'Document Crawler',
-    nameShort: 'Crawler',
+    id: 'control-pipeline',
+    name: 'Control Pipeline',
+    nameShort: 'Pipeline',
     layer: 'backend',
     tech: 'Python / FastAPI',
     port: 8098,
     url: 'https://macmini:8098',
-    container: 'bp-compliance-document-crawler',
-    description: 'Dokument-Analyse (PDF, DOCX, XLSX, PPTX), Gap-Analyse, IPFS-Archivierung.',
-    descriptionLong: 'Der Document Crawler nimmt hochgeladene Dokumente (PDF, DOCX, XLSX, PPTX) entgegen, extrahiert deren Inhalt und fuehrt eine Gap-Analyse gegen bestehende Compliance-Anforderungen durch. Dafuer leitet er die Textinhalte an den AI Compliance SDK weiter, der die semantische Analyse uebernimmt. Abgeschlossene Dokumente koennen ueber den DSMS-Service dezentral auf IPFS archiviert werden.',
-    dbTables: [],
-    ragCollections: [],
-    apiEndpoints: [
-      'POST /analyze',
-      'POST /gap-analysis',
-      'POST /archive',
+    container: 'bp-core-control-pipeline',
+    description: 'RAG-zu-Controls Pipeline: Control Generation, Pass 0a/0b, Ontology, Dedup, Dependency Engine, Applicability.',
+    descriptionLong: 'Die Control Pipeline ist das Herzsttueck der automatisierten Compliance-Control-Generierung. Sie verarbeitet ~105.000 RAG-Chunks aus EU/DE-Regulierungen in 6 Phasen: (1) RAG Ingestion, (2) 7-Stufen Control Generation (Lizenz-Gate + Claude LLM), (3) Pass 0a Obligation Extraction (~181k Obligations), (4) Pass 0b Atomic Composition (MCP-taugliche Controls mit assertion/pass_criteria/fail_criteria), (5) Embedding-basierte Deduplizierung mit LLM-Verifikation, (6) Dependency Engine (5 Typen: supersedes, prerequisite, compensating_control, scope_exclusion, conditional_requirement) mit automatischer Generierung via Ontology, Pattern-Regeln und Domain Packs (DSGVO, AI Act, CRA, Security, Arbeitsrecht). 126+ Tests, alle bestanden.',
+    dbTables: [
+      'canonical_controls', 'obligation_candidates', 'control_parent_links',
+      'control_dependencies', 'control_evaluation_results',
+      'canonical_processed_chunks', 'canonical_generation_jobs',
+      'control_dedup_reviews', 'control_patterns',
     ],
-    dependsOn: ['ai-compliance-sdk', 'dsms'],
+    ragCollections: [
+      'bp_compliance_gesetze', 'bp_compliance_datenschutz',
+      'bp_compliance_ce', 'bp_dsfa_corpus', 'bp_legal_templates',
+    ],
+    apiEndpoints: [
+      'POST /v1/canonical/generate',
+      'GET /v1/canonical/controls',
+      'POST /v1/canonical/controls/applicable',
+      'POST /v1/canonical/generate/submit-pass0b',
+      'POST /v1/canonical/generate/process-batch',
+      'GET /v1/canonical/generate/quality-metrics',
+      'POST /v1/dependencies/generate',
+      'POST /v1/dependencies/evaluate',
+      'GET /v1/dependencies/graph',
+      'POST /v1/document-compliance/required',
+    ],
+    dependsOn: ['postgresql', 'qdrant', 'ollama'],
   },
   {
     id: 'compliance-tts',
@@ -383,7 +398,7 @@ export const ARCH_EDGES: ArchEdge[] = [
   // Frontend → Backend
   { source: 'admin-compliance', target: 'backend-compliance', label: 'REST API' },
   { source: 'admin-compliance', target: 'ai-compliance-sdk', label: 'REST API' },
-  { source: 'admin-compliance', target: 'document-crawler', label: 'REST API' },
+  { source: 'admin-compliance', target: 'control-pipeline', label: 'REST API' },
 
   // Backend → Infrastructure
   { source: 'backend-compliance', target: 'postgresql', label: 'SQLAlchemy' },
@@ -392,12 +407,9 @@ export const ARCH_EDGES: ArchEdge[] = [
   { source: 'ai-compliance-sdk', target: 'ollama', label: 'LLM Inference' },
   { source: 'ai-compliance-sdk', target: 'postgresql', label: 'GORM' },
   { source: 'compliance-tts', target: 'minio', label: 'Audio/Video' },
-
-  // Backend → Backend
-  { source: 'document-crawler', target: 'ai-compliance-sdk', label: 'LLM Gateway' },
-
-  // Backend → Data Sovereignty
-  { source: 'document-crawler', target: 'dsms', label: 'IPFS Archive' },
+  { source: 'control-pipeline', target: 'postgresql', label: 'SQLAlchemy' },
+  { source: 'control-pipeline', target: 'qdrant', label: 'Embedding + Dedup' },
+  { source: 'control-pipeline', target: 'ollama', label: 'LLM Dedup (qwen3.5)' },
 ]
 
 // =============================================================================