For relatively prime $m$ and $n$, we can split up an fraction with denominator $mn$ into fractions with denominators $m$ and $n$: $mx+ny = 1$ for some integers $m$ and $n$, so
$$
\frac{1}{mn} = \frac{y}{m} + \frac{x}{n}.
$$
The ambiguity in the choice of numerators on the right side is reflected in the ambiguity in the choice of integers $x$ and $y$ making $mx+ny = 1$: if $x_0$ and $y_0$ are one solution then all others are $x=x_0+nk$ and $y=y_0-nk$ for $k\in \mathbf Z$.
Example. To break up $1/15$, we need to solve $3x+5y=1$. One choice is $x=2$ and $y=-1$, so
$$
\frac{1}{15} = \frac{-1}{3} + \frac{2}{5}.
$$
The general choices for $x$ and $y$ are $x=2+5k$ and $y=-1-3k$, so
$$
\frac{1}{15} = \frac{-1-3k}{3} + \frac{2+5k}{5}.
$$
In particular, it is impossible to make such a decomposition of $1/15$ with both numerators nonnegative. (That’s also obvious since $1/3$ and $1/5$ both exceed $1/15$.)
The same result applies with a denominator that is a product of relatively prime nonconstant polynomials $f(t)$ and $g(t)$: we can write $uf + vg = 1$ for some polynomials $u(t)$ and $v(t)$, so
$$
\frac{1}{fg} = \frac{v}{f} + \frac{u}{g}.
$$
Polynomials are different from integers in an important way: there are unique minimal choices of $u$ and $v$: there are unique $u$ and $v$ such that $\deg u < \deg g$ and $\deg v < \deg f$. Moreover, that uniqueness also holds if we want to write a polynomial $h$ as $uf+vg$ when $\deg h < \deg(fg)$. That leads to
$$
\frac{h}{fg} = \frac{v}{f} + \frac{u}{g},
$$
which is why the partial fraction decomposition of a proper ratio (numerator of smaller degree than denominator) leads to a sum of parts that involves only proper ratios. There is no analogue of this for rational numbers because there is no canonical way to solve $mx+ny=1$ in integers. More specifically, the set of all possible remainders under polynomial division by a fixed polynomial is an additive group: $\{r : \deg r < d\} \cup \{0\}$ is an additive group. There is nothing like that in the integers. If we did not insist on low-degree numerators in a partial fraction decomposition of rational functions then the partial fraction decomposition of rational functions would not be unique!
Something special about Euclidean domains of the form $F[t]$ for a field $F$ is that it has unique quotients and remainders for its Euclidean function. That is not true for the integers or the Gaussian integers with their Euclidean functions $|n|$ and ${\rm N}(\alpha)$. This sounds surprising for integers, where we are taught that there is a unique quotient and remainder! But be careful: that happens when we require $a=bq+r$ with $0\leq r < |b|$, which does not involve the Euclidean function on $r$. If we use $|r|$ in place of $r$ to measure the size of $r$ then $r$ is not unique since it could be negative: $10 = 4\cdot 2 + 2 = 4\cdot 3-2$. The only Euclidean domains other fields with unique quotients and remainders are the polynomial rings $F[t]$. Bill has helpfully provided a link that will direct you to proofs of this in one of his comments above.