Fix typesupport loading

[Karsten1987]

The change to rcpputils::SharedLibrary in #322 introduced a regression to the converter logic.

copied from #322

I haven't paid enough attention while reviewing this PR, but I think it's really bad practice to pass in a shared_pointer by reference (std::shared_pointerrcpputils::SharedLibrary & library) and somewhat blindly overwrite it by calling std::make_shared on it. Also, this assumes that the pointer is empty, which isn't checked here either.
The reason why it fails in the end is that the same pointer is being passed in to load multiple typesupports, which are stored in add_topics. Later on, when using these typesupport symbols to create a new introspection message, these pointers have dangled.

This PR is fixing this by extending the ConverterTypeSupport struct with the shared library. It is currently using a shared pointer to it, even though I don't understand why this is needed and a reference isn't sufficient. Maybe @ahcorde can give some more insights here.

This is needed to unblock ros2/rosbag2_bag_v2#28

[Karsten1987]

• Linux
• Linux-aarch64
• macOS
• Windows

[Karsten1987]

• Linux
• Linux-aarch64
• macOS
• Windows

[Karsten1987]

• Linux
• Linux-aarch64
• macOS
• Windows

[Karsten1987]

That is up for review.

I am still a bit concerned about this as every introspection message is now strongly coupled with the rcpputils::SharedLibrary in order to be cleaned up correctly. However, this would need a bigger round of code review and in order to get the beta going and the rosbag2_bag_v2_plugin unblocked, this should be good to go.

[Karsten1987]

@emersonknapp @zmichaels11 friendly ping for review