How to construct the radial component of the momentum operator?
You would have to use the fact that the momentum operator in position space is $\vec{p} = -i\hbar\vec{\nabla}$ and use the definition of the gradient operator in spherical coordinates:
$$\vec{\nabla} = \hat{r}\frac{\partial}{\partial r} + \hat{\theta}\frac{1}{r}\frac{\partial}{\partial\theta} + \hat{\phi}\frac{1}{r\sin\theta}\frac{\partial}{\partial\phi}$$
So the radial component of momentum is
$$p_r = -i\hbar\hat{r}\frac{\partial}{\partial r}$$
However: after a bit of investigation prompted by the comments, I found that in practice this is not used very much. It's more useful to have an operator $p_r'$ that satisfies
$$-\frac{\hbar^2}{2m}\nabla^2 R(r) = \frac{p_r'^2}{2m} R(r)$$
This lets you write the radial component of the time-independent Schrödinger equation as
$$\biggl(\frac{p_r'^2}{2m} + V(r)\biggr)R(r) = E R(r)$$
The action of the radial component of the Laplacian in 3D is
$$\nabla^2 R(r) = \frac{1}{r^2}\frac{\partial}{\partial r}\biggl(r^2\frac{\partial R(r)}{\partial r}\biggr)$$
and if you solve for the operator $p'_r$ that satisfies the definition above, you wind up with
$$p'_r = -i\hbar\biggl(\frac{\partial}{\partial r} + \frac{1}{r}\biggr)$$
This is called the "radial momentum operator." Strictly speaking, it is different from the "radial component of the momentum operator," which is, by definition, $p_r$ as I wrote it above, although I wouldn't be surprised to find people mixing up the terminology relatively often.
I was able to figure it out, so here goes the clarification for the record.
Classically $$p_r=\hat{D}_r = \frac{\hat{r}}{r} \cdot\hat{p} = \frac{\hbar}{i}\frac{\partial}{\partial r}$$
However $\hat{D}_r$ is not hermitian. Consider the adjoint $$\hat{D}_r^\dagger= \hat{p}\cdot\frac{\hat{r}}{r} =\frac{\hbar}{i} \left ( \frac{\partial}{\partial r}+\frac{2}{r} \right ) $$
Now we know from linear algebra how to construct a hermitian operator from an operator and its adjoint: $$\hat{p}_r = \frac{\hat{D}_r^\dagger+\hat{D}_r}{2}=\frac{\hbar}{i} \left ( \frac{\partial}{\partial r}+\frac{1}{r} \right ) $$
And btw, for those who followed my initial question, don't do the following calculational mistake that I committed:
$$\hat{p}_r = \frac{1}{2}\left(\frac{1}{r}(\vec{r}\cdot\vec{p})+(\vec{p}\cdot\vec{r})\frac{1}{r}\right)\neq \frac{1}{2}\frac{1}{r}(\vec{r}\cdot\vec{p}+\vec{p}\cdot\vec{r})$$
Even in one dimension the operator $p_r=-i\partial_r$ on the half line $r>0$ has deficiency indices $(0,1)$. There is thus no way to define it it as a self-adjoint operator. In practical terms this abstract mathematical statement means that there is no set of boundary conditions thta we can impose on the wavefunction $\psi(r)$ that lead to a complete set of eigenfunctions for $p_r$. For example, integration by parts to prove formal hermiticity requires that $\psi(0)=0$ but all potential eigenfunctions are of the form $\psi_k(r)=\exp\{ikr\}$ for some real or complex $k$, and no value of $k$ can make $\psi_k(0)$ be zero.
Since one needs eigenfunctions and eignvalues to assign a probability to the outcome of a measurement of $p_r$ this means that $p_r$ is not an ``observable.''