Get Issue Intake Images

The screenshots themselves, base64, so Rooster can LOOK at them.

He used to be told he could not see inside an attachment, and said so to reporters -- asking people to describe a picture he was already holding. Separate from the context endpoint because image bytes are large and every other caller of that endpoint wants text. Kept for whatever else reads the raw attachments (e.g. an eventual UI); the vision pass itself reads through describe_issue_intake_images below, which fetches these same bytes server-side rather than round-tripping them through a caller.