@Jaeyoung-Lim Thanks for the review!
I implemented most of your suggestions and left some comments.

After discussing the implementation with @bresch I made some major changes to how the intersection point is calculated. It is now done using vectors which eliminates many of the edgecases I had to consider.
Also to keep everything consistent with @Jaeyoung-Lim's input:

If you return a zero, there is no way outside of the library to detect whether if the input you have provided is valid or not.
It is actually a positive feature if something breaks down the line. It means that you need to handle these cases explicitly for your runtime application, rather than concealing it in the library. The way it is implemented now, an invalid position estimate will result in aligning the vehicle into zero heading, which is a behavior that is unrelated to the failure.

the algorithm now returns NAN if any of the inputs are invalid rather than using the bearing to the current waypoint as a fallback.

rebased on main and squashed