[JTC] Return INVALID_GOAL error code for rejected trajectories (#700)

[AdvityaCode]

Problem
When the Joint Trajectory Controller rejects an invalid trajectory (empty points, wrong joint names, non-monotonic timestamps, etc.), the action client receives an opaque REJECT response with no structured feedback. The only signal available to the client is a log line in the controller's terminal — which is not a viable error-handling strategy for automated systems.

This is a limitation of action protocol (tracked upstream at ros2/rclc#271): GoalResponse::REJECT provides no mechanism to set error_code or error_string on the result.

Closes #700.

Proposed Solution
This is my suggested approach — I'm open to feedback and happy to revise the implementation. It is just what I thought would be the best approach.

The workaround discussed in #699 is to always accept goals at the action protocol level, then immediately abort with a populated error_code and error_string if validation fails. I followed that direction here, but I wasn't sure on some of the design choices so please let me know if any of these should be done differently:

• goal_received_callback now unconditionally returns ACCEPT_AND_EXECUTE. Validation is deferred to goal_accepted_callback, where a structured result can be returned to the client.
• goal_accepted_callback performs two checks before processing a goal:
  - Whether the controller is active (PRIMARY_STATE_INACTIVE → abort with INVALID_GOAL)
  - Whether the trajectory message is valid (all existing checks → abort with INVALID_GOAL + descriptive error_string)
• validate_trajectory_msg and validate_trajectory_point_field now return std::string instead of bool. An empty string means valid; a non-empty string is the human-readable rejection reason. Each path still logs via RCLCPP_ERROR for operator visibility.
• The topic-based trajectory interface (~/joint_trajectory) is unaffected in behavior.

I wasn't certain whether changing the return type of validate_trajectory_msg to std::string was the right call, or whether an out-parameter or a separate struct would be preferred. Happy to change this if there's a style the project prefers.

Tests

• All existing validate_trajectory_msg unit tests updated for the new return type.
• Two new parameterized action tests added:
  • test_goal_aborted_invalid_trajectory_empty_points
  • test_goal_aborted_invalid_trajectory_wrong_joint_names

[christophfroehlich]

Thank you for picking this.

We will have to discuss this in the next PMC meeting if the accept-and-immediately-reject pattern is something everyone is happy with.

I propose that we use tl::expected (until std::expected is available) instead of returning an empty string.

[christophfroehlich]

what do you think of using tl-expected here? Similar to the validators in https://github.com/PickNikRobotics/generate_parameter_library?tab=readme-ov-file#custom-validator-functions

[AdvityaCode]

I agree with using tl-expected! Just updated the pr: switched validate_trajectory_msg and validate_trajectory_point_field to return tl::expected<void, std::string>. The tl_expected header was already available via validate_jtc_parameters.hpp so no new dependencies were needed. Let me know your thoughts and if there's anything else you'd like changed.