Ok, let me see if I can shed light on some of the questions raised.
"What fails if I try the construction with non-smooth stuff?" This question is a bit unspecific, a lot of things can fail, depending on what you want. For instance, if you want higher Chow groups to come out of the construction, you need homotopy invariance which is false for non-smooth schemes.
For the construction of categories of motives starting from a site of (possibly non-smooth) schemes, you would usually use some topology like $h$-, $qfh$- or $cdh$-topology. In these topologies, schemes will be locally smooth because some sort of resolution of singularities gives you coverings. In some cases (best with rational coefficients), there are comparison results to the Nisnevich topology.
The Lefschetz motive questions:
If you take Voevodsky's embedding from Chow motives into $DM_{gm}$, the image of the Lefschetz motive is the Tate motive $\mathbb{Z}(1)[2]$. This is the reduced motive of $\mathbb{G}_m$ or $\mathbb{P}^1$, appropriately shifted, i.e., either $\ker\left(C_\ast\mathbb{Z}_{tr}(\mathbb{G}_{m,S})\to C_\ast\mathbb{Z}_{tr}(S)\right)$.
The embedding from Chow motives to Voevodsky motives is given by mapping a variety $X$ to $M(X)=C_\ast\mathbb{Z}_{tr}(X)$. Since Voevodsky's category is idempotent complete and has transfers, this assignment factors through Chow motives. Hence, the embedding takes a Chow motive $(X,p)$ to the summand of $M(X)$ given by the projector $p$. I should think that this is discussed in the red book (Friedlander, Suslin, Voevodsky: Cycles, transfers and motivic homology theories), at least the book contains the theorem that Chow motives embed as full subcategory of Voevodsky motives.
I would say that there is no geometric meaning to $\mathbb{Z}(-1)$ (unless you consider virtual bundles as something geometric). It is simply constructed as the $\otimes$-inverse of $\mathbb{Z}(1)$, in the same way as spectra arise from CW-complexes by inverting $S^1$ w.r.t. the smash product. Somehow, all this technical stabilization business is necessary because there is no geometric meaning to $\mathbb{Z}(-1)$ (although this is more like a philosophical belief).
The construction of motives over more general bases is currently being studied. Check out the works of Joseph Ayoub, cf. his homepage where you can find his papers. His ICM paper gives a nice overview of the state of the art. I guess the easiest construction of motives over arbitrary bases is via étale sheaves made $\mathbb{A}^1$-invariant and $\mathbb{P}^1$-stable. For the resulting category, a full six functor formalism is available...
There are also constructions using Nisnevich sheaves with transfers, cf. the work of Cisinski-Déglise, but the construction of these categories is a bit complicated in full generality because of difficulties with the intersection theory. With rational coefficients, these categories can be compared to the étale version; with integral coefficients, there are problems with the localization triangle.
Representability of smash products: I do not know of any results, but I would guess that the smash product is more likely not represented by a scheme. I am more confident saying that the smash product is most likely not represented by a smooth scheme: there is a conjecture of Asok stating that the spheres $S^{p,q}\cong S^p\wedge\mathbb{G}_m^{\wedge q}$ are not $\mathbb{A}^1$-homotopic to smooth schemes unless $p=2q$ or $p=2q-1$. Admittedly, that's a bit different from the corresponding question about motives... The right way to think about the smash product is: it's the same thing as in topology, same categorical properties etc.
Finally, I would again say that there is no geometric meaning to shifting $\mathbb{Z}(q)$. The topological analogues are exactly the Eilenberg-Mac Lane spaces, one should think of $\mathbb{Z}(q)[n]$ as the motive of $K(\mathbb{Z}(q),n)$. In topology as well, there is no real geometric relation between $K(\mathbb{Z},n)$ and $K(\mathbb{Z},n-1)$...
I would also like to point out that $\mathbb{Z}_{tr}\left(\bigwedge^n\mathbb{G}_m\right)$ does not really have a meaning, because $\mathbb{Z}_{tr}$ takes a scheme and turns it into a sheaf with transfers - you should take $\mathbb{Z}(1)$ and then take its smash powers, not the other way round.