is it true that any numerical polynomial , i.e. $f(t)\in \mathbb Q[t], f(n)\in\mathbb Z\ \forall n\in\mathbb Z\ $ can be presented as Hilbert polynomial of some variety?
This paper (Proposition 1.3) gives a necessary and sufficient condition for a polynomial to be the Hilbert polynomial of a projective scheme. 


Steven Landsburg's answer completely solves the problem for schemes. I notice that you asked for a variety, which is more restrictive. For example, $2t+2$ is the Hilbert polynomial of a pair of skew lines, but it is not the Hilbert polynomial of any irreducible variety. Such a variety would have to be a degree $2$ curve, by looking at the leading coefficient, and this curve would have to have genus $1$ by RiemmanRoch, a contradiction. Similarly, $2 \binom{t+2}{2} 1 = t^2+3t+1$ is the Hilbert polynomial of two $\mathbb{P}^2$'s which meet transversely as a point and I believe that it is not the Hilbert polynomial of any irreducible variety. 

