Refactor CollectionModel/builder and introduce read-only property interfaces (#13699)

## Changes

* **Extract `ConfigureVectorPropertyEmbedding()`**: Deduplicated
embedding resolution logic that was repeated across
`ProcessTypeProperties` and `ProcessRecordDefinition`.
* **Replace `IRecordCreator` with `Func<object>`**: Removed
`IRecordCreator` interface and two implementing classes
(`ActivatorBasedRecordCreator`, `DynamicRecordCreator`), replacing them
with simple lambdas.
* **Rename `TemporaryStorageName` → `SerializedKeyName`**: Moved from
`PropertyModel` base to `KeyPropertyModel` where it belongs (only used
by CosmosNoSql for key property JSON serializer name remapping).
* **Delegate-based property accessors**: Replaced virtual
`GetValueAsObject`/`SetValueAsObject` overrides with delegate fields
(`_getter`/`_setter`), configured via
`ConfigurePocoAccessors()`/`ConfigureDynamicAccessors()`. Converted
null-coalescing throws to `Debug.Assert`.
* **Improved xmldoc** on `VectorPropertyModel.EmbeddingType` explaining
the `[AllowNull]` invariant.
* **Introduce read-only interfaces** (`IPropertyModel`,
`IKeyPropertyModel`, `IDataPropertyModel`, `IVectorPropertyModel`):
`CollectionModel` now exposes `IReadOnlyList<IVectorPropertyModel>`
etc., giving providers an immutable view post-build. All provider code
updated to consume interface types.

Note that the last is a non-trivial breaking change for providers (not
users), which I think is fine (note that the provider-facing APIs are
flagged as [Experimental]).

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: roji

dotnet/Directory.Packages.props

@@ -102,7 +102,7 @@
     <PackageVersion Include="PdfPig" Version="0.1.13" />
     <PackageVersion Include="Pinecone.Client" Version="3.1.0" />
     <PackageVersion Include="Prompty.Core" Version="0.2.3-beta" />
-    <PackageVersion Include="Scriban" Version="6.6.0" />
+    <PackageVersion Include="Scriban" Version="7.0.3" />
     <PackageVersion Include="PuppeteerSharp" Version="20.2.5" />
     <PackageVersion Include="System.Diagnostics.DiagnosticSource" Version="10.0.2" />
     <PackageVersion Include="System.IdentityModel.Tokens.Jwt" Version="8.15.0" />

dotnet/src/InternalUtilities/connectors/Memory/MongoDB/MongoModelBuilder.cs

@@ -27,17 +27,14 @@ internal class MongoModelBuilder() : CollectionModelBuilder(s_validationOptions)
         UsesExternalSerializer = true,
     };
 
-    [RequiresUnreferencedCode("Traverses the CLR type's properties with reflection, so not compatible with trimming")]
-    protected override void ProcessTypeProperties(Type type, VectorStoreCollectionDefinition? definition)
+    protected override void ProcessProperty(PropertyInfo? clrProperty, VectorStoreProperty? definitionProperty, Type? type)
     {
-        base.ProcessTypeProperties(type, definition);
+        base.ProcessProperty(clrProperty, definitionProperty, type);
 
-        foreach (var property in this.Properties)
+        if (clrProperty?.GetCustomAttribute<BsonElementAttribute>() is { } bsonElementAttribute
+            && this.PropertyMap.TryGetValue(clrProperty.Name, out var property))
         {
-            if (property.PropertyInfo?.GetCustomAttribute<BsonElementAttribute>() is { } bsonElementAttribute)
-            {
-                property.StorageName = bsonElementAttribute.ElementName;
-            }
+            property.StorageName = bsonElementAttribute.ElementName;
         }
     }
 

dotnet/src/VectorData/CosmosNoSql/CosmosNoSqlMapper.cs

@@ -41,8 +41,8 @@ public JsonObject MapFromDataToStorageModel(TRecord dataModel, int recordIndex,
 
         // The key property in Azure CosmosDB NoSQL is always named 'id'.
         // But the external JSON serializer used just above isn't aware of that, and will produce a JSON object with another name, taking into
-        // account e.g. naming policies. TemporaryStorageName gets populated in the model builder - containing that name - once VectorStoreModelBuildingOptions.ReservedKeyPropertyName is set
-        RenameJsonProperty(jsonObject, this._keyProperty.TemporaryStorageName!, CosmosNoSqlConstants.ReservedKeyPropertyName);
+        // account e.g. naming policies. SerializedKeyName gets populated in the model builder - containing that name - once VectorStoreModelBuildingOptions.ReservedKeyPropertyName is set
+        RenameJsonProperty(jsonObject, this._keyProperty.SerializedKeyName!, CosmosNoSqlConstants.ReservedKeyPropertyName);
 
         // Go over the vector properties; inject any generated embeddings to overwrite the JSON serialized above.
         // Also, for Embedding<T> properties we also need to overwrite with a simple array (since Embedding<T> gets serialized as a complex object).
@@ -116,7 +116,7 @@ public JsonObject MapFromDataToStorageModel(TRecord dataModel, int recordIndex,
     public TRecord MapFromStorageToDataModel(JsonObject storageModel, bool includeVectors)
     {
         // See above comment.
-        RenameJsonProperty(storageModel, CosmosNoSqlConstants.ReservedKeyPropertyName, this._keyProperty.TemporaryStorageName!);
+        RenameJsonProperty(storageModel, CosmosNoSqlConstants.ReservedKeyPropertyName, this._keyProperty.SerializedKeyName!);
 
         foreach (var vectorProperty in this._model.VectorProperties)
         {

dotnet/src/VectorData/SqliteVec/SqlitePropertyMapping.cs

@@ -62,7 +62,7 @@ public static List<SqliteColumn> GetColumns(IReadOnlyList<PropertyModel> propert
             else
             {
                 // The Key column in included in both Vector and Data tables.
-                Debug.Assert(property is KeyPropertyModel, "property is VectorStoreRecordKeyPropertyModel");
+                Debug.Assert(property is KeyPropertyModel, "property is not a KeyPropertyModel");
 
                 propertyType = GetStorageDataPropertyType(property);
                 isPrimary = true;

dotnet/src/VectorData/VectorData.Abstractions/ProviderServices/CollectionJsonModelBuilder.cs

@@ -89,13 +89,11 @@ protected override void Customize()
 
             if (keyPropertyWithReservedName)
             {
-                // Somewhat hacky:
                 // Some providers (Weaviate, Cosmos NoSQL) have a fixed, reserved storage name for keys (id), and at the same time use an external
                 // JSON serializer to serialize the entire user POCO. Since the serializer is unaware of the reserved storage name, it will produce
                 // a storage name as usual, based on the .NET property's name, possibly with a naming policy applied to it. The connector then needs
                 // to look that up and replace with the reserved name.
-                // So we store the policy-transformed name, as StorageName contains the reserved name.
-                property.TemporaryStorageName = storageName;
+                ((KeyPropertyModel)property).SerializedKeyName = storageName;
             }
             else
             {

dotnet/src/VectorData/VectorData.Abstractions/ProviderServices/CollectionModel.cs

@@ -19,7 +19,7 @@ namespace Microsoft.Extensions.VectorData.ProviderServices;
 public sealed class CollectionModel
 {
     private readonly Type _recordType;
-    private readonly IRecordCreator _recordCreator;
+    private readonly Func<object> _recordFactory;
 
     private KeyPropertyModel? _singleKeyProperty;
     private VectorPropertyModel? _singleVectorProperty;
@@ -57,14 +57,14 @@ public sealed class CollectionModel
 
     internal CollectionModel(
         Type recordType,
-        IRecordCreator recordCreator,
+        Func<object> recordFactory,
         IReadOnlyList<KeyPropertyModel> keyProperties,
         IReadOnlyList<DataPropertyModel> dataProperties,
         IReadOnlyList<VectorPropertyModel> vectorProperties,
         IReadOnlyDictionary<string, PropertyModel> propertyMap)
     {
         this._recordType = recordType;
-        this._recordCreator = recordCreator;
+        this._recordFactory = recordFactory;
 
         this.KeyProperties = keyProperties;
         this.DataProperties = dataProperties;
@@ -97,7 +97,7 @@ public TRecord CreateRecord<TRecord>()
     {
         Debug.Assert(typeof(TRecord) == this._recordType, "Type mismatch between record type and model type.");
 
-        return this._recordCreator.Create<TRecord>();
+        return (TRecord)this._recordFactory();
     }
 
     /// <summary>