fix(image-generation): let LLM deliver images via message tool instead of runtime media attachment

The runtime media-attachment mechanism was broken for streaming channels
(e.g. WebSocket): the _streamed flag caused _send_once to skip the final
OutboundMessage that carried generated media, so images were never delivered.

Rather than adding complex coordination between streaming and media delivery,
delegate image delivery to the LLM: after generate_image returns artifact
paths, the next_step prompt now instructs the LLM to call the message tool
with the paths in the media parameter. This works uniformly across all
channels, streaming or not.

Remove generated_media from TurnContext, _assemble_outbound, and _state_save.
Update prompts in identity.md, SKILL.md, message tool description, and
artifacts.py to reflect the new flow.

nanobot/skills/image-generation/SKILL.md

@@ -15,7 +15,7 @@ If the `generate_image` tool is not available in the current tool list, tell the
 - Image editing: pass the saved artifact path or user image path in `reference_images`.
 - Iterative edits in the same conversation: prefer the most recent generated image artifact if the user says things like "make it brighter", "change the background", or "try another version".
 - Ambiguous edits: ask a short clarifying question if multiple recent images could be the target.
-- In the current chat, do not call `message` just to announce or resend generated images. The runtime attaches images from `generate_image` to the final assistant reply automatically.
+- After generating images, call the `message` tool with the artifact paths in the `media` parameter to deliver them to the user.
 
 ## Prompt Rules