.Net: Consider ImageContent and AudioContent inherit from BinaryContent

[SergeyMenshykh]

@matthewbolanos please provide details of the issue that would be fixed by the change.

[Krzysztof318]

@SergeyMenshykh I don't think it will be good. BinaryContent.Data is required but ImageContent.Data and AudioContent.Data are optional, so casting these two to BinaryContent would give inconsistent behavior. Following this way, we should create a DataKernelContent class with Uri and BinaryData parameters and inherit from it.

[matthewbolanos]

This feature should be worked on in conjunction with https://github.com/microsoft/semantic-kernel/issues/5263 (which should address @Krzysztof318's concerns around BinaryContent.Data being required). Because BinaryContent could/should be hydrated with either a URL or a byte array, it should likely be renamed to FileContent.

The main scenario that we'd like to support is the following...

Context

• Today, the BinaryContent class is predominantly used for the Assistants API
• The Assistants API supports several different file types: https://platform.openai.com/docs/assistants/tools/supported-files
• Most notably, it supports image types like jpg and png
• In the future, it may also support other types like audio and video

Scenario

• The user makes a request to the Assistant API
• The Assistant API returns back a collection with an image file and audio file
• The developer wants to easily save all the files to disk without needing to type cast the contents (i.e., use the base class)
• The developer then wants to take the image and run it through another model (e.g., image-to-text to get a description for it for a11y). Instead of taking the binary of the file and creating a new ImageContent with it, the user should simply be able to cast it to the correct type.

[matthewbolanos]

As a test of the new hierarchy, the following POC should be possible:

1. Ask an agent built with the Assistant API with Code Interpreter to create an 1) image, 2) audio file, and 3) word file
2. Get the collection of files from the ChatMessageContent (today they come in via the AnnotationContent, so there needs to be a way to easily turn this into BinaryContent/FileContent; perhaps AnnotationContent is merged with BinaryContent/FileContent and the data like Quote can be stored in a metadata field)
3. Save all the files to disk using the parent class (e.g., BinaryContent or FileContent)
4. Type caste the image to ImageContent to perform a text-to-image task
5. Type caste the audio to AudioContent to perform audio-to-text task