promote_type involving NAtype doesn't seem right

[garborg]

(As mentioned a comment near the promote_type definition, but deserves an issue.)

Came up on julia-users: https://groups.google.com/forum/#!topic/julia-users/piCb9m7lUmE

Breaking vcat(dfs) when a column of NAtype and another type T are vcat'ed:

x = similar(promote_type(...), ...) # is of type `T`
x[i] = NA # fails to convert to type T

Not expecting immediate action because of the other work going on, but putting it up here.

[johnmyleswhite]

Never been too happy with how we defined promote_type, but agree that it's a low priority when we're shifting to Nullable.