There is no inconsistency. Indeed, when using the full polarizability in the random phase approximation
$\chi^{\text{RPA}}(\mathbf{q},\omega)=\dfrac{\chi^{\text{RPA}}_0(\mathbf{q},\omega)}{1-V(q)\chi^{\text{RPA}}_0(\mathbf{q},\omega)}$
you see that the imaginary part of the polarizability has been renormalized when taken into account the [renormalized] electron-electron interactions because $V(q)$ is no longer the bare Coulomb interaction. Note that, from the previous expression, if we set $\text{Im}\,\chi_0(\mathbf{q},\omega)=-i\delta$
$-\text{Im} \,\chi(\mathbf{q},\omega)=\pi \delta(1-V(q)\text{Re}\,\chi_0(\mathbf{q},\omega))$
The fact that the imaginary part of the polarizability function $\text{Im} \,\chi(\mathbf{q},\omega)$ is non-zero is related to the damping of the excitations and it is known as Landau damping.
It is possible to understand this linking $\text{Im} \,\chi(\mathbf{q},\omega)$ to the conductivity of the electron gas. Using the definition of the conductivity and the continuity equation we can show that
$e^2 \text{Im} \,\chi(\mathbf{q},\omega) = -\dfrac{1}{\omega} \mathbf{q}\cdot[\text{Re}\,\bar\sigma(\mathbf{q},\omega)]\cdot \mathbf{q} $
The real part of the conductivity is related to dissipation in the system [Joule heating] when a current $\mathbf{J}$ is flowing.
Best Answer
First, the Thomas-Fermi screening is a semiclassical static theory which assumes that the total potential $\phi(\mathbf{r})$ varies slowly in the scale of the Fermi length $l_{\text{F}}$, the chemical potential $\mu$ is constant and that $T$ is low. In principle, it does not rely on linear response theory.
The condition of slowly varying potential is a general condition of validity of semiclassical models. Physically, if the particle [electron] is represented by a wave packet, what is tellying us is that all the waves in the wavepacket will see the same potential and the particle will suffer [or enjoy!] a force as if it was point-like ["classical"] because such potentials gives rise to ordinary forces in the equation of motion describing the evolution of the position and wavevector of the packet. The wavepacket must have a well-defined wavevector on scale of the Brillouin zone [thus $\Delta k \simeq k_{\text{F}}$] and therefore can be spread in the real space over many primitive cells.
Mathematically, the assumption that your potential is a slowly varying function of the position implies that the theory is not valid for $|\mathbf{q}| \gg k_{\text{F}}$ [and therefore for $|\mathbf{r}| \ll l_{\text{F}}$].
On the other hand, the static Lindhard dielectric function is a fully quantum treatment of the problem and it is valid for all the ranges of $\mathbf{q}$. It includes, in the limit $\mathbf{q} \rightarrow 0$, the linearized Thomas-Fermi dielectric function. It only assumes linear response, that is, the induced density of charge is proportional to the total potential $\phi(\mathbf{r})$.
Note also that the Lindhard treatment is far more general than the Thomas-Fermi in the sense that it can describe both dynamic and static screening.