Improve docx parsing, chunk ordering, and DB config

- Add null check for Document in DocxContentDecoder to prevent exceptions.
- Set DocumentChunk.Id to auto-generate in ApplicationDbContext.
- Order vector search results by cosine similarity for relevance.

Author: marcominerva

SqlDatabaseVectorSearch/ContentDecoders/DocxContentDecoder.cs

@@ -14,7 +14,7 @@ public Task<IEnumerable<Chunk>> DecodeAsync(Stream stream, string contentType, C
         // Open a Word document for read-only access.
         using var document = WordprocessingDocument.Open(stream, false);
 
-        var body = document.MainDocumentPart?.Document.Body;
+        var body = document.MainDocumentPart?.Document?.Body;
         var content = new StringBuilder();
 
         foreach (var p in body?.Descendants<Paragraph>() ?? [])

SqlDatabaseVectorSearch/Data/ApplicationDbContext.cs

@@ -34,7 +34,7 @@ protected override void OnModelCreating(ModelBuilder modelBuilder)
         modelBuilder.Entity<DocumentChunk>(entity =>
         {
             entity.ToTable("DocumentChunks");
-            entity.HasKey(e => e.Id);
+            entity.HasKey(e => e.Id);            
 
             entity.Property(e => e.Id).ValueGeneratedOnAdd();
             entity.Property(e => e.Content).IsRequired();

SqlDatabaseVectorSearch/Services/VectorSearchService.cs

@@ -152,7 +152,7 @@ public async IAsyncEnumerable<Response> AskStreamingAsync(Question question, boo
         var questionEmbedding = await embeddingGenerator.GenerateVectorAsync(reformulatedQuestion.Text!, cancellationToken: cancellationToken);
         var embeddingVector = new SqlVector<float>(questionEmbedding);
 
-        var chunks = await dbContext.DocumentChunks.Include(c => c.Document)                    
+        var chunks = await dbContext.DocumentChunks.Include(c => c.Document)
                     .OrderBy(c => EF.Functions.VectorDistance("cosine", c.Embedding, embeddingVector))
                     .Take(appSettings.MaxRelevantChunks)
                     .ToListAsync(cancellationToken);