Consider the coefficient of $t$ in LHS: $$\displaystyle\sum_{i=1}^n x_i\sum_{j\neq i}-\frac{1}{x_j-x_i}=\sum_{i=1}^n\sum_{j\neq i}\frac{x_i}{x_i-x_j}=\sum_{i<j}\frac{x_i}{x_i-x_j}+\frac{x_j}{x_j-x_i}=\sum_{i<j}1={n\choose 2}$$
However, I don't know how to prove the coefficient of $t^s$ is $0$ for $s\geq 2$.
The identity comes from:
Show $$ \sum_{i=1}^n x_i \, \Delta(x_1, \cdots, x_i + t, \cdots, x_n) = \left( \sum_{i=1}^n x_i + \binom{n}{2} t \right) \Delta(x_1, \cdots, x_n) $$ where $$ \Delta(x_1, \cdots, x_n) := \prod_{1 \leq i < j \leq n} (x_i - x_j) $$
If we cancel $\Delta(x_1, \cdots, x_n)$ from both sides, then we get the identity in the title.
$\Delta(x_1, \cdots, x_n)$ can be represented by the determinant of corresponding Vandermonde matrix, I don't know whether it's helpful.