Support Microsoft MAI image generation in Copilot agent workflows

## Problem

Copilot agents can help build applications, but generating visual assets with Microsoft MAI models currently requires a separate workflow.

An agent should be able to generate an image during a coding conversation, show the result in chat, and optionally save it into the project. This should work independently of whether the selected conversational model supports native image generation.

## Proposal

Add a built-in **Generate Image** tool backed by Microsoft MAI image models.

## Expected behavior

- Available in normal Copilot Agent chat and the local Agents Window.
- Generates PNG images from text prompts.
- Displays results through the existing image preview, carousel, and Save actions.
- Preserves generated previews when the conversation is reopened.
- Optionally creates a new project file in the originating workspace or session worktree, without overwriting existing files.
- Uses secure credential storage and existing tool approval controls.
- Reports cancellation, authentication, quota, content-policy, and save failures clearly.
- Hides with disabled AI features.

## Initial scope

Start with a user-configured Microsoft Foundry endpoint and API key. Image generation is billed to the configured Azure resource.

A dedicated product endpoint and authentication using an existing signed-in identity can follow separately. Image editing, remote agent hosts, and headless execution are outside the initial scope.