Breaking this portion out from SERVER-81848:

We can assume that zero-init is good enough for some types that aren't trivial init, such as boost::optional (safer since we control the definition) and std::unique_ptr<T> with no custom deleter (slightly more risky since the definition is provided by the toolchain, however it is hard to imagine a practical implementation where zero filling isn't correct, and our tests would explode if it wasn't)