[Tobias-Fischer]

Fixes #430. ros2_control 6.12 (ros-controls/ros2_control#3610) removed the pointer-based StateInterface/CommandInterface constructors and export_state_interfaces()/export_command_interfaces(), so kortex_driver no longer builds on Rolling.

The port keeps the driver's logic and its interface lists:

• The former export loops now describe (prefix, name, &member) bindings. Interfaces the URDF declares (per joint: command position, state position/velocity/effort; gripper: command position, state position/velocity) come from the base class. The ones it doesn't (per-joint velocity/effort commands, set_gripper_max_velocity/set_gripper_max_effort, tcp/twist.*, reset_fault/command, reset_fault/async_success, reset_fault/internal_fault) are exported as unlisted interfaces with the same names, all double, so controllers and configurations need no changes.
• Every member behind an interface is written only by the control loop, so the sync is simple: write() first copies the command handles into the members, read() ends by copying the state members into their handles, and wherever the driver itself resets commands (perform_command_mode_switch, on_activate, the fault reset in write()) it writes the handles too. After export, each command handle starts at its member's initial value (e.g. the gripper's max velocity/effort).
• Two small fixes on the way: arm_commands_velocities_ is reset with std::fill instead of a six-element literal (which shrank it to six entries on the 7-DoF arm, and the sync loops index every joint), and the reset_fault members are initialised.

Formatted with clang-format 18 (the pre-commit version); no new cppcheck findings.

Tested:

• Builds against Rolling's hardware_interface 6.12.0 (RoboStack, linux-64), on the 0.2.5 release with the CMake part of Kortex API & Rolling Binary Build Updates #415, which is the same C++ change as this PR; kortex_bringup and kinova_gen3_7dof_robotiq_2f_85_moveit_config build on top.
• An offline check that loads the real xacro URDFs, runs on_init up to the robot connection, and lists the exported interfaces: for gen3 7-DoF with and without the gripper and gen3 6-DoF, the names and types match exactly what 0.2.5 exported, and the state/command sync works in both directions.
• Not tested on hardware or through controller_manager; that's the part that needs someone with a robot.

Unrelated, but noticed while testing on Rolling: robotiq_description 0.0.1 rejects the mock_sensor_commands argument that kortex_description passes to the gripper xacro.

🤖 Claude (AI assistant), on behalf of @Tobias-Fischer

🤖 Generated with Claude Code