Feature: Llava executor

[SignalRT]

This is just a preview, the code it's not clean or finished. But I think It's enough to talk about the implementation.

Introduction to the example

• Changes in the Interactive Executor to add an additional constructor to enable the support of vision models.
• An example using the Interactive Executor to create and initialize the llava model.

This is just a raw capture of the example working:

Things to talk about:

• The interface to the executors. I just create some properties in ILLavaExecutor (I know there are some duplicate property) basically to identify if it's a vision model and the current Image. This was the easiest approach to get an example working. I would like to get your suggestions on the integration on ILLamaExecutor interface. Especially I didn´t review the implications in the Chat or Semantic Kernel integrations, to allow to support that kind of models in all the layers.

[AsakusaRinne]

LGTM, please convert it from draft to formal PR if it's ready to merge.

[SignalRT]

I will solve the current conflicts today and change the PR to formal PR.

[SignalRT]

The latest changes are:

• Change the example to make a "preview" in the console of the image. Just to have an idea of the file loaded.
• Allow to introduce image files with the path in the console (in the example).
• Theoretically to be able to manage more than a file.

Notes: The example is just tested with llava-v1.6-mistral. The prompt could change if the model used is a different one.

This is an execution of the example: