feat: Semantische Suche über Chat-History (memory_search)

- Lokales Embedding-Model (all-MiniLM-L6-v2, ONNX, CPU)
- EmbeddingService: WordPiece-Tokenizer + Mean Pooling (384 Dim)
- VectorSearchService: SQLite-basierte Vektor-Speicherung + Cosine Similarity
- EmbeddingBackgroundService: Backfill bestehender Messages + Live-Queue
- Neues Tool memory_search: Semantische Suche über alle Sessions (inkl. archivierte)
- ~9000 Messages embedded, Suche über komplette Chat-History

Services/EmbeddingBackgroundService.cs

@@ -0,0 +1,148 @@
+using System;
+using System.Collections.Concurrent;
+using System.Collections.Generic;
+using System.Threading;
+using System.Threading.Tasks;
+using Dapper;
+using Microsoft.Data.Sqlite;
+using Microsoft.Extensions.Hosting;
+
+namespace Gott.Services;
+
+/// <summary>
+/// Background-Service der Messages embedded:
+/// 1. Beim Start: Backfill aller bestehenden Text-Messages
+/// 2. Laufend: Neue Messages über Queue verarbeiten
+/// </summary>
+public class EmbeddingBackgroundService : BackgroundService
+{
+    readonly VectorSearchService _vectorSearch;
+    readonly EmbeddingService _embedding;
+    readonly string _dbConnectionString;
+    readonly ConcurrentQueue<EmbeddingJob> _queue = new();
+    readonly SemaphoreSlim _signal = new(0);
+
+    public EmbeddingBackgroundService(VectorSearchService vectorSearch, EmbeddingService embedding, string dbPath)
+    {
+        _vectorSearch = vectorSearch;
+        _embedding = embedding;
+        _dbConnectionString = $"Data Source={dbPath}";
+    }
+
+    /// <summary>Enqueue eine neue Message zum Embedden.</summary>
+    public void Enqueue(long messageId, string chainId, string sessionId, string agentId, string role, string text)
+    {
+        _queue.Enqueue(new EmbeddingJob(messageId, chainId, sessionId, agentId, role, text));
+        _signal.Release();
+    }
+
+    protected override async Task ExecuteAsync(CancellationToken ct)
+    {
+        if (!_embedding.IsAvailable)
+        {
+            Console.WriteLine("[EmbeddingBG] Embedding model not available – service disabled.");
+            return;
+        }
+
+        // Kurz warten bis App vollständig gestartet
+        await Task.Delay(3000, ct);
+
+        // Backfill bestehender Messages
+        await BackfillAsync(ct);
+
+        // Live-Embedding neuer Messages
+        Console.WriteLine("[EmbeddingBG] Waiting for new messages...");
+        while (!ct.IsCancellationRequested)
+        {
+            try
+            {
+                await _signal.WaitAsync(ct);
+                while (_queue.TryDequeue(out var job))
+                {
+                    if (ct.IsCancellationRequested) break;
+                    try
+                    {
+                        await _vectorSearch.EmbedMessageAsync(job.MessageId, job.ChainId, job.SessionId,
+                            job.AgentId, job.Role, job.Text);
+                    }
+                    catch (Exception ex)
+                    {
+                        Console.WriteLine($"[EmbeddingBG] Error embedding message {job.MessageId}: {ex.Message}");
+                    }
+                }
+            }
+            catch (OperationCanceledException) { break; }
+        }
+    }
+
+    async Task BackfillAsync(CancellationToken ct)
+    {
+        Console.WriteLine("[EmbeddingBG] Starting backfill of existing messages...");
+
+        var existingIds = await _vectorSearch.GetEmbeddedMessageIdsAsync();
+        Console.WriteLine($"[EmbeddingBG] Already embedded: {existingIds.Count} messages");
+
+        using var conn = new SqliteConnection(_dbConnectionString);
+        await conn.OpenAsync(ct);
+
+        // Alle text-Messages laden (user + assistant, inkl. archivierte Sessions)
+        var messages = await conn.QueryAsync<BackfillRow>(@"
+            SELECT m.id, m.chain_id, c.session_id, c.agent_id, m.role, m.content, m.type
+            FROM messages m
+            JOIN chains c ON m.chain_id = c.chain_id
+            WHERE m.type = 'text'
+              AND m.role IN ('user', 'assistant')
+              AND m.content IS NOT NULL
+              AND LENGTH(m.content) >= 10
+              AND m.deleted_on IS NULL
+            ORDER BY m.id ASC
+        ");
+
+        int total = 0;
+        int embedded = 0;
+        int skipped = 0;
+
+        foreach (var msg in messages)
+        {
+            if (ct.IsCancellationRequested) break;
+            total++;
+
+            if (existingIds.Contains(msg.Id))
+            {
+                skipped++;
+                continue;
+            }
+
+            try
+            {
+                // SessionId kann auch aus archive kommen
+                var sessionId = msg.SessionId ?? "";
+                await _vectorSearch.EmbedMessageAsync(msg.Id, msg.ChainId, sessionId,
+                    msg.AgentId, msg.Role, msg.Content);
+                embedded++;
+
+                if (embedded % 100 == 0)
+                    Console.WriteLine($"[EmbeddingBG] Backfill progress: {embedded} embedded, {skipped} skipped, {total} total");
+            }
+            catch (Exception ex)
+            {
+                Console.WriteLine($"[EmbeddingBG] Error backfilling message {msg.Id}: {ex.Message}");
+            }
+        }
+
+        Console.WriteLine($"[EmbeddingBG] Backfill complete: {embedded} new embeddings, {skipped} skipped, {total} total messages.");
+    }
+
+    record EmbeddingJob(long MessageId, string ChainId, string SessionId, string AgentId, string Role, string Text);
+
+    class BackfillRow
+    {
+        public long Id { get; set; }
+        public string ChainId { get; set; }
+        public string SessionId { get; set; }
+        public string AgentId { get; set; }
+        public string Role { get; set; }
+        public string Content { get; set; }
+        public string Type { get; set; }
+    }
+}