Declare the hardware_interface dependency explicitly

[nbbrooks]

Fixes #36.

Why

Both controllers use hardware_interface directly:

	include	usage
picknik_reset_fault_controller	src/…:26 #include "hardware_interface/loaned_command_interface.hpp"	using hardware_interface::LoanedCommandInterface;
picknik_twist_controller	src/…:31 same header	same

Neither declared hardware_interface in package.xml, and neither find_packaged it. They relied on it arriving transitively:

picknik_*_controller -> controller_interface -> hardware_interface

As @Plumezz noted in #36, this is not a build break today. It's a metadata problem: a dependency we use directly is invisible to rosdep and to anyone reading the manifest, and it would break silently if controller_interface ever stopped re-exporting it.

Two additions to the original report: the gap is on main as well as humble, and it affects both packages, not just picknik_reset_fault_controller.

How

For each package:

• <depend>hardware_interface</depend> in package.xml
• find_package(hardware_interface REQUIRED)
• hardware_interface added to THIS_PACKAGE_INCLUDE_DEPENDS, so it's re-exported to consumers alongside the other direct deps
• ${hardware_interface_TARGETS} in target_link_libraries, matching the ${..._TARGETS} style already used for controller_interface

On the last point — I checked that hardware_interface_TARGETS resolves to hardware_interface::hardware_interface only. mock_components is in a separate export set (export_mock_components) and is not pulled in, so this doesn't widen the link beyond what's intended.

8 lines added, nothing removed.

Verification

Built both packages from a clean workspace in a ros:lyrical-ros-base container: rosdep install resolves the newly-declared dependency, and colcon build succeeds for both.

Separate finding: 12 ignored set_value() return values

Not fixed here, since it needs a behaviour decision rather than a metadata one — but the build surfaces it, so worth recording.

LoanedCommandInterface::set_value() is [[nodiscard]] on lyrical and returns false if it cannot acquire the lock within max_tries (default 10). Both controllers discard it at every call site — 12 -Wunused-result warnings, 6 per package:

• picknik_reset_fault_controller.cpp: 87, 88, 113, 114, 122, 123
• picknik_twist_controller.cpp: 148–153

warning: ignoring return value of 'bool hardware_interface::LoanedCommandInterface::set_value(
  const T&, unsigned int) [with T = double]', declared with attribute 'nodiscard' [-Wunused-result]

So both controllers can currently drop a command write silently. Note the read path was already modernised — get_optional() with value_or(...) — so this is the write half of the same API migration left unfinished. Worth its own issue; happy to file it and take it if you want.

Sequencing

Branched off main, so until #38 merges this PR only exercises the currently-active workflows. Once #38 is in I'll rebase so it gets the full matrix, including the blocking lyrical-main job.

The same fix is wanted on the humble branch, which has the identical gap in both packages. Say the word and I'll open that as a companion PR.