Use cuRobo IK for GearAssembly reset grasp pose

[karanchahal-nv]

Summary

• Replace the GearAssembly reset grasp-pose handcrafted Jacobian/DLS IK loop with batched cuRobo IK.
• Cache the cuRobo solver and fixed-size full-env goal/seed buffers so reset calls can reuse CUDA graph shapes.
• Configure the UR10e GearAssembly envs with the cuRobo IK defaults used during profiling: ur10e.yml, 1 seed, 3 grad iters, 1 Newton iter, full-batch CUDA graph solve.

Motivation

Profiling showed set_robot_to_grasp_pose reset work was a major reset-side cost. The profiling harness validated the same cuRobo path before this PR.

Validation

• python3 -m compileall -q source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/deploy/mdp/events.py source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/deploy/gear_assembly/config/ur_10e/joint_pos_env_cfg.py
• /home/horde/IsaacLab/isaaclab.sh -p - <<'PY' ... import smoke ... PY

Notes

• This targets the GearAssembly deploy task branch, not upstream IsaacLab main.
• cuRobo must be installed/available in the runtime environment, as in the existing profiling setup.