Contour tree performance improvements (ContourTreeMesh)

Various performance improvements to ContourTreeMesh implementation

• Avoid using STL functions (std::merge, std::inplace_merge and std::unique) by replacing them with worklets and custom implementations.
• Use ScanExtended to avoid having to explicitly set first array value
• Use ArrayGetValue instead of ReadPortal for getting single array values in the control environment without having to transfer the full array.
• Refactor to use more vtk-m functions (fancy arrays, existing conversion functions) to avoid custom worklets.
• Improve comments and readability.