Improve compile times

[tzcnt]

Possible optimizations:

Template hoisting:

• move common functionality of templates (that doesn't depend on any template parameters) into base class
• move nested template types out
• replace template parameter types in the above with void*, and static_cast<T*> in the derived template class
• base classes are declared in .hpp and defined in .ipp (#ifdef TMC_IMPL)

Replace std::move / std::forward with static_cast -> remove header

Diagnose other issues with clang -ftime-trace