Image generation doesn't work in air gapped systems

[emerzon]

Today, a image generation requires a second call to an external Azure Endpoint in order to retrieve the generated image. This won't work in air-gapped system as they will not have access to this external URL.

Ideally the image generation should rely on receiving the image inline in b64 format, as part of the model response.

An initial PR was crafted at #1861, however a lot has changed in how images are handled internally and this PR no longer works, and would need to be adapted.