[waveshare-ugv] Wire localized maps into Leash visualization, memory, and waypoints

[superposition]

Context

Once the concrete UGV produces a stable map and pose, the existing Leash tools should consume them through generic interfaces before physical autonomous execution is enabled.

Depends on #161 and #166.

Scope

• Surface map identity, pose, covariance, occupancy/costmap, path, and localization health through telemetry, visualization, MCP, recording, and replay.
• Bind spatial-memory entries and saved waypoints to the active map identity/frame.
• Allow perception results to be tagged into map-frame spatial memory when localization is healthy.
• Handle map replacement, reload, and localization loss without silently reusing stale coordinates.
• Keep physical planner and patrol execution disabled.

Definition of done

• HTTP/WebSocket telemetry, visualization, MCP observe, recording, and replay report the same map identity and localized pose.
• A map-frame location and observed object can be tagged, listed, queried, and replayed through spatial memory.
• A saved waypoint can be created, listed, updated, and restored after map reload without moving the robot.
• Map replacement or localization loss marks incompatible memory/waypoints stale or rejects their use explicitly.
• Occupancy/costmap and current pose render correctly in the operator visualization.
• Localization health and covariance are visible to operators and agents.
• Automated fixtures cover healthy, stale, lost, map-replaced, and recovered states.
• Physical goal and patrol starts remain rejected by default.
• Schema, replay, MCP HTTP, visualization, and complete no-hardware smoke gates pass.