Extract and cleanup serialization/deserialization writing logic

[AlexanderSher]

Model generation in DPG Grow-Up will require serialization / deserialization code generated in a variety of places. We can try to use existing serialization / deserialization writers, but those would require some improvements.

Current state:

• SerializationWriter contains entry point for serialization/deserialization writing for SchemaObjectType. It writes method signatures and deserialization logic for discriminators
• RequestWriterHelpers writes some parts of the serialization inline in the CreateXxxRequest method
• JsonCodeWriterExtensions and XmlCodeWriterExtensions contains core serialization / deserialization writing logic.

Existing problems:

• JsonPropertySerialization and JsonAdditionalPropertiesSerialization references ObjectTypeProperty, which makes JSON serialization bound to ObjectType, no random object can be serialized. Same problem with XmlObjectAttributeSerialization.
• Methods in JsonCodeWriterExtensions and XmlCodeWriterExtensions use chain of callbacks (XmlCodeWriterExtensions,ToDeserializeCall - callback of callback, Action<CodeWriter, CodeWriterDelegate>), which makes it very hard to debug or modify this code.
• Some variables are created without CodeWriterDeclaration which can result in name conflict

Expected result:

• Make object graph that represents data for JSON and XML serialization self-contained, so it can be created from any set of objects or constructed from scratch for testing purposes
• Move all graph construction logic from writer into builder (e.g. CollectProperties is not writer logic)
• Get rid of callbacks where possible (ideally - get rid of all callbacks).
• Ensure we control for name conflicts

[annelo-msft]

We can start on this item once v1 GA work is completed.