feat: unified_search + ai_proactive plugins with Ollama Cloud DeepSeek V4

- unified_search: Hybride Suche (PostgreSQL FTS + pgvector + RRF Fusion)
  - 5 Search Providers (Contact, Company, Mail, File, Event)
  - KI Query Understanding (Fuzzy, Facetten via LiteLLM)
  - DMS Text-Extraction (PDF, DOCX, XLSX, PPTX)
  - Embedding Pipeline (ollama/nomic-embed-text, 768 Dim)
  - Background Jobs für Indexierung
  - Plugin-basierte Provider Registry

- ai_proactive: Proaktiver KI-Agent
  - Context-Tracking (Frontend → Backend → Event Bus)
  - Proactive Engine mit LLM Suggestion-Generierung
  - SSE Real-time Push an Frontend
  - 6 AI Tools für Tool Registry
  - Rate-Limiting + User Settings
  - Deep Analysis Background Jobs

- Frontend Integration:
  - useAIContext Hook, SuggestionSidebar, SuggestionBadge
  - ProactiveAISettings Page, Search API Client
  - Globale Suche auf neue API umgestellt

- Tests: test_unified_search.py + test_ai_proactive.py (alle bestanden)
- Config: Ollama Cloud DeepSeek V4 als Default, konfigurierbar
- Dependencies: PyMuPDF, python-docx, python-pptx, pgvector
- Bugfixes: notification type_key length, migration IF NOT EXISTS

app/plugins/builtins/unified_search/embedding.py

@@ -0,0 +1,131 @@
+"""Embedding pipeline using LiteLLM with configurable Ollama Cloud provider."""
+
+from __future__ import annotations
+
+import os
+import logging
+import uuid
+from typing import TYPE_CHECKING
+
+import litellm
+
+if TYPE_CHECKING:
+    from sqlalchemy.ext.asyncio import AsyncSession
+
+logger = logging.getLogger(__name__)
+
+MAX_INPUT_CHARS = 8000
+
+DEFAULT_EMBEDDING_MODEL = os.environ.get('SEARCH_EMBEDDING_MODEL', 'ollama/nomic-embed-text')
+OLLAMA_API_KEY = os.environ.get('API_KEY_OLLAMA_CLOUD', '')
+
+
+async def generate_embedding(text: str, model: str | None = None) -> list[float]:
+    """Generate a single embedding via LiteLLM.
+
+    Args:
+        text: Input text (truncated to 8000 chars).
+        model: Embedding model name (default: ollama/nomic-embed-text).
+
+    Returns:
+        Embedding vector as list of floats.
+    """
+    model = model or DEFAULT_EMBEDDING_MODEL
+    truncated = text[:MAX_INPUT_CHARS]
+    try:
+        response = await litellm.aembedding(
+            model=model,
+            input=truncated,
+            api_key=OLLAMA_API_KEY,
+        )
+        return response.data[0]["embedding"]
+    except Exception:
+        logger.exception("Failed to generate embedding")
+        return []
+
+
+async def generate_embeddings_batch(
+    texts: list[str], model: str | None = None
+) -> list[list[float]]:
+    """Generate embeddings for multiple texts in a single API call.
+
+    Args:
+        texts: List of input texts.
+        model: Embedding model name (default: ollama/nomic-embed-text).
+
+    Returns:
+        List of embedding vectors.
+    """
+    model = model or DEFAULT_EMBEDDING_MODEL
+    truncated = [t[:MAX_INPUT_CHARS] for t in texts]
+    try:
+        response = await litellm.aembedding(
+            model=model,
+            input=truncated,
+            api_key=OLLAMA_API_KEY,
+        )
+        return [d["embedding"] for d in response.data]
+    except Exception:
+        logger.exception("Failed to generate batch embeddings")
+        return [[] for _ in texts]
+
+
+async def index_entity(
+    entity_type: str,
+    entity_id: uuid.UUID,
+    tenant_id: uuid.UUID,
+    db: AsyncSession,
+) -> bool:
+    """Generate and store embedding for a single entity.
+
+    Uses the provider registry to get embedding text, generates embedding,
+    and updates the entity's embedding column.
+
+    Returns True on success, False on failure.
+    """
+    from app.plugins.builtins.unified_search.provider_registry import get_search_registry
+
+    registry = get_search_registry()
+    provider = registry.get(entity_type)
+    if provider is None:
+        logger.warning("No provider for entity_type=%s", entity_type)
+        return False
+
+    try:
+        text = await provider.get_embedding_text(db, entity_id, tenant_id)
+        if not text.strip():
+            logger.debug("Empty embedding text for %s/%s", entity_type, entity_id)
+            return False
+
+        embedding = await generate_embedding(text)
+        if not embedding:
+            return False
+
+        # Update the entity's embedding column
+        from sqlalchemy import text as sql_text
+
+        table_map = {
+            "contact": "contacts",
+            "company": "companies",
+            "mail": "mails",
+            "file": "files",
+            "event": "calendar_entries",
+        }
+        table = table_map.get(entity_type)
+        if not table:
+            logger.warning("Unknown entity_type=%s for embedding storage", entity_type)
+            return False
+
+        sql = sql_text(
+            f"UPDATE {table} SET embedding = cast(:emb AS vector) "
+            f"WHERE id = :eid AND tenant_id = :tid"
+        )
+        await db.execute(
+            sql,
+            {"emb": str(embedding), "eid": entity_id, "tid": tenant_id},
+        )
+        await db.commit()
+        return True
+    except Exception:
+        logger.exception("Failed to index entity %s/%s", entity_type, entity_id)
+        return False