The RBF Gram matrix

Write the TypeScript function rbfGram with signature

$\texttt{rbfGram}(X,\gamma)\ \longmapsto\ K\in\mathbb R^{m\times m},\qquad K_{ij}=\exp\!\big(-\gamma\,\lVert x_i-x_j\rVert^2\big)=\exp\!\Big(-\gamma\sum_{t=1}^{d}(x_{i,t}-x_{j,t})^2\Big),$

where row $i$ of X is the data point $x_i\in\mathbb R^d$ and gamma is $\gamma>0$. Return a plain array of $m$ rows of $m$ numbers each. The norm is squared: no square root is taken.

Visible test three-points: $X=\{(0,0),(1,0),(0,2)\}$ and $\gamma=0.5$. The squared distances are $\lVert x_0-x_1\rVert^2=1$, $\lVert x_0-x_2\rVert^2=4$ and $\lVert x_1-x_2\rVert^2=1+4=5$, so

$K=\begin{pmatrix}1&e^{-0.5}&e^{-2}\\ e^{-0.5}&1&e^{-2.5}\\ e^{-2}&e^{-2.5}&1\end{pmatrix}\approx\begin{pmatrix}1&0.6065306597&0.1353352832\\0.6065306597&1&0.0820849986\\0.1353352832&0.0820849986&1\end{pmatrix},$

compared entrywise with absolute tolerance $10^{-9}$.

Hidden tests (arguments are sent to your browser, expected values are not): three data sets of $m=20$ points in $[-2,2]^2$ with $\gamma=0.5$, $1$ and $2$. For each set: (values) every entry within $10^{-9}$ of the reference matrix; (symmetric) $\max_{ij}\lvert K_{ij}-K_{ji}\rvert\le10^{-12}$; (psd) the smallest eigenvalue of $K$, computed by the engine's Hermitian eigensolver, is at least $-10^{-10}$.

Use Math.exp; do not import anything.