[v3] Add constexpr specifiers to std::shared_ptr, std::unique_ptr, etc

[Benjamin Kosnik]

> To be clear: those tests were obviously incorrect, because unique_ptr
> and shared_ptr are not literal types.

Agreed that unique_ptr and shared_ptr are not literal types.

However, those cons/constexpr.cc tests aren't testing if they are
literal types... I don't see how these tests were obviously incorrect.
Can you explain this to me please?

Shouldn't 

constexpr std::unique_ptr<int> a;

work now, if indeed constexpr constexpr() is implemented?

The only use I see right now for a default-constructed constexpr
unique_ptr is to have a non-dymamically-initialized check for a "null"
unique_ptr. So I'm a bit curious about this design decision. 

Anyway.

I agree with your point about the constexpr data tests: they are not
really testing constexpr data members at the moment, just static. Duly
noted. I am hoping to improve those with the scan-assembler trick Jason
suggested. Hang on, we'll get there.

-benjamin