The question asks whether it is possible to mathematically prove the speed of propagation of gravity waves without linearization.

It does not yet appear possible. Please let me explain.

Einstein's GR postulates the equality of a curvature tensor $R_{ij}-R g_{ij}$ with a mass-energy quantity of the form $\kappa T_{ij}$ where $T$ is a term representing mass-energy. That is $R_{ij}-Rg_{ij}=\kappa T_{ij}$

The equality demands:

Hypothesis (i) that the LHS term of Einstein's GR equation  $R_{ij}-Rg_{ij}$ is an invariant tensor, and has vanishing tensor divergence $div(R_{ij}-Rg_{ij})=0$. (And this is true, as verified by calculations). 

Hypothesis (ii):  that the RHS mass-energy quantity $T_{ij}$ be a tensor quantity, and that $T_{ij}$ have well-defined tensor divergence which vanishes identically.  

The GR equation appears to be well formed only if (ii) is satisfied.

In Einstein's formulation of GR (1916) an expression for the mass energy $T$ was given, but which failed to satisfy (ii). The error is an improper index contraction. The master of the absolute differential calculus, Levi-Civita corrected Einstein's original expression $t^\alpha_\sigma$ for the so-called "energy components" $t^\alpha_\sigma$ of the gravitational fields. Levi-Civita's form of GR equations yields proper tensor divergences $div T$ *if* $T$ is tensorial. See Ch.XI, SS 20-25 in Levi-Civita's "Absolute Differential Calculus". However Levi-Civita's appears to confuse *mass* with *matter*, invoking the continuity equation of incompressible fluid flow as parallel for the energy density using $e=mc^2=\rho c^2$. But we see no reason why *mass* needs satisfy a continuity equation. For we acknowledge that *matter* is neither created nor destroyed, but *mass* is mutable. This is controversial, relating to Newton's definition of *mass* as a measure of *matter*.

The tensoriality of $T$ is not readily established, being dependant on the physical model used.

As far as Einstein's original expression $t=t^\alpha_\sigma$, 
P.A.M. Dirac said ``in general, gravitational energy cannot be localized. The best we can do is use a pseudotensor...which gives us approximate information about gravitational energy, which in some special cases can be accurate." (See Dirac's book "General Theory of Relativity"). 

A.S. Eddington similarly concluded the nonpossibility, writing: ``If coordinates are chosen so as to satisfy a certain condition which has no very clear geometrical importance, the speed [of gravity waves] is that of light; if the coordinates are slightly different the speed is altogether different from that of light. The result stands or falls by the choice of coordinates, ...". (See Eddington, The Mathematical Theory of Relativity, S 57). 

The key point -- as realized by Einstein, Eddington, Dirac, Hoyle, Abrams, even Crothers -- is that Einstein's so-called "gravitational energy tensor" is *not* a tensor at all! To quote Einstein: "The quantities $t^\alpha_\sigma$ we call the 'energy components' of the gravitational field,..., it is to be noted that $t^\alpha_\sigma$ is not a tensor". (See Einstein's ``The Foundation of the General Relativity, 1916, S.15). 

Einstein noted well that $t$ is not a tensor, but is invariant under *linear unimodular* change of coordinates. This is elaborated in many excellent articles by E.Norton (see his articles on General Covariance and Einstein's Point-Coincidence Argument, and the long documented struggles which Einstein had in developing satisfactory covariant equations. 

A further difficult is that Einstein apparently discovers the conservation of gravitational energy by evaluating the coordinate divergence(!) of $t$ and finding $\partial t^\alpha_\sigma / \partial x_\alpha=0$. He says "This equation expresses the law of conservation of momentum and of energy for the gravitational field." (Ibid) However, the vanishing of a coordinate divergence of a nontensor object is not a covariant object except (in this case) for observers who share the same volume form, i.e. unimodular linear change of coordinates change of coordinates. 

Dirac says further, "Let us consider the energy of these waves. Oweing to the pseudo-tensor not being a real tensor, we do not get, in general, a clear result independant of the coordinate system. But there is one special case in which we do get a clear result; namely when the waves are all moving in the same direction", (Ibid). 

So as of yet, the answer to the OP's question appears Negative.