The server should return better error codes when generating OpenAI images with unacceptable parameters

[KallynGowdy]

Currently, when calling ai.generateImage() with a parameter that isn't accepted by the selected model, the server will simply return a server_error code. The should be able to handle HTTP 400 Bad Request responses from the model provider (openai, google, etc.), and return a proper response to the user.

Examples of parameters which can cause this issue:

• width=123 with dall-e-2 and dall-e-3
• steps < 10 with stable-diffusion-xl-1024-v1-0