Investigate the benefits of optimizations with trivial types in memory algorithms

[dmitriy-sobolev]

Why do we even think that having assignment (via fill) instead of calling value-constructor over the memory is anyhow better?

In other words, why are we trying to avoid placement new so hard?

Originally posted by @rarutyun in #2292 (comment)