Reorganize URDF packages, fix mesh resolution, enhance gravity comp controller

[Yang-Ci]

Summary

This PR reorganizes the URDF packages, fixes a mesh resolution bug, and enhances the gravity compensation controller with new tuning parameters.

Changes

URDF Package Reorganization

• Removed obsolete packages urdf/00-arm-rs_asm-v3/ (old RS) and urdf/reBot-DevArm_fixend_description/ (old DM). Neither was referenced by any code outside the urdf/ directory.
• Added urdf/DM/ with ReBot_Arm_DM.urdf featuring separated visual/collision meshes, multi-material colored visuals, and gripper finger definitions (prismatic joints with mimic coupling).
• Added urdf/RS/ with ReBot_Arm_RS.urdf and organized mesh directories (visual/, shared/, mujoco_collision/).
• Updated urdf_path in config/rebotarm_dm.yaml, config/rebotarm_rs.yaml, and tools/gravity_calibration/*.py to point to the new package paths.

Mesh Resolution Fix

• _resolve_urdf() in robot_model.py returned a single package directory derived from the URDF path. This broke when a URDF mixed two mesh-path conventions: ../meshes/... (relative to URDF dir) and meshes/... (relative to package root).
• Now returns both the URDF directory and its parent as package_dirs, so Pinocchio resolves either convention without per-URDF workarounds.
• Applied the same fix to example/sim/visualizer.py.

Gravity Compensation Controller Enhancements

• tau_scale: per-joint gravity torque scaling to correct model-vs-hardware mass/inertia bias.
• transition_duration: smooth gain fade-in from stiff hold to compliant gravity-comp gains, preventing sudden stiffness loss at startup.
• joint_direction: per-joint sign correction for motor/URDF axis mismatches.
• hold_kp/hold_kd: stiff position hold gains; end() now maintains the current pose with stiff gains + gravity feedforward instead of simply disconnecting.
• Added gravity_compensation config sections to both DM and RS YAML files with calibrated tau_scale values.
• Updated example scripts (9, 10) to use the new controller parameters.

Config Tuning

• Switched default hardware in rebotarm.yaml from DM to RS.
• Tuned POS_VEL pos_kp for joints 4-6 in rebotarm_dm.yaml (50 -> 70/60/60).

Additional Fixes

• MIT control example: accept either 6 arm-only values or 7 arm-plus-gripper values. A 6-value command preserves the current gripper target, while a 7-value command updates it. This behavior is driven by the configured group sizes and works identically for DM and RS.
• FK simulation: keep the documented 6-value arm input while padding the remaining URDF gripper/finger coordinates to zero internally, so the command no longer incorrectly requires 8 values.
• Interaction cleanup: support inline comments, reject malformed numeric input safely, and ensure MIT control exits through disconnect on Ctrl+C or EOF.

[Yang-Ci]

Superseded by #28, which combines the original URDF and gravity-controller changes with the new YAML profiles and robust RobStride velocity-lock implementation.