agent encode for PoseStamped

[leshy]

if my skill returns some posestamped (and other stuff potentially)

PoseStamped needs to be rendered to an agent (maybe as text description relative to the robot body) so agent knows "this pose is 5 meters to my 3 o clock" and more then that - agent should be able to ask to navigate to it, rotate towards it etc (I guess can just support global coordinate nav)

This way I can ask the agent "what is the coolest place you saw last 5 minutes" this is a memory2 query - probably semantic query, returns images and poses, agent should be able to look at the image, be able to describe it, and navigate to pose