Consider simplifications in the implementation of `uninitialized_*` algorithms

[akukanov]

There are at least two questions worth investigating:

• Do the if constexpr checks in the implementation to "optimize" for types with trivial operations provide any practical performance improvements?
• Are all variations of copy, move. fill etc. bricks/functors used in the implementation really necessary (or could it all be refactored to reduce the number of those and maximize re-use)?