Add a CommandPtr with an internal unique_ptr to enable not needing to move the underlying classes, which is error-prone due to the potential for lambda captures.
- Remove use of std::set. The only place std::set was actually used was in ParallelRaceGroup,
but this was of minimal utility as ParallelRaceGroup checked for duplicate subsystem
requirements, so it would be very unusual to end up with duplicate commands
in any case; replaced it with a vector.
- Remove use of std::unordered_map except for SelectCommand. Replaced with vector.
- Use pImpl idiom for CommandScheduler
- Minimize include files (remove unnecessary ones)
- Reformat include file order for consistency
