Wanted to clarify one potentially confusing difference between Śniatycki's and Prof. Figueroa-O'Farrill's notation: $\beta$ is actually the *local representative* of the connection $2\pi i~\alpha$ with respect to the non-vanishing local section $\sigma$, i.e., its pullback under $\sigma$
$$
\beta = 2\pi i~\sigma^*\alpha~.
$$
Then Prof. Figueroa-O'Farrill's equation
$
Ds = (df_s + \beta f_s) \sigma
$
is equivalent to Śniatycki's equation (3.17),
$
\nabla_X\sigma = 2\pi i~\sigma^*\alpha(X)~\sigma.
$

The function $h$ mentioned above is the logarithm of (the positive-valued) $\langle\sigma,\sigma\rangle$
$$
h = \ln\langle\sigma,\sigma\rangle = \sigma^{*}g,
$$
where $g$ is the function $p\mapsto\ln\langle p,p\rangle$ on the non-vanishing part of the line bundle $L$.

Hence $dh=\beta+\overline{\beta}$ becomes
$
d(\sigma^* g) = 2\pi i(\sigma^* \alpha - \overline{\sigma^* \alpha})~,
$
or, since $\sigma$ is arbitrary and pullback commutes with exterior derivative
$$
dg = 2\pi i(\alpha-\overline{\alpha}).
$$

Conversely, if such a $g$ exists, the quadratic form $e^g$ (necessarily positive since $g$ is necessarily real) will give your desired pairing via the polarization identity, and this pairing will be unique up to a positive constant for connected base space.

The geometric interpretation of this condition is as follows: if $p\in L-\textrm{zero section}, u\in \mathbb{R}\simeq\mathfrak{u}(1)$, and $\eta_u(p) = \frac{d}{dt}e^{2\pi i ut}p\big\vert_{t=0}$ is the fiber $U(1)$-generator at $p$ (which by definition satisfies $\alpha(\eta_u(p))=u$ in Śniatycki's conventions), then $dg = 2\pi i(\alpha-\overline{\alpha})$ gives
$$
\eta_u(p)g = 0
$$
or in other words $g(e^{2\pi i u}p) = g(p)$, as would be expected if $g$ is to equal $\ln\langle\cdot,\cdot\rangle$ for some pairing. Correspondingly, the fiber radial generators are given by $\eta_{-iv},~v\in\mathbb{R}$, and
$$
\eta_{-iv}(p)g = 4\pi v
$$
or $g(e^{2\pi v}p) = 4\pi v+g(p)$ for $p\in L-\textrm{zero section}, v\in\mathbb{R}$, again as would be expected.

The best place to learn the full theory is still Kostant's original "Quantization and Unitary Representations", published in Lecture Notes in Mathematics 170. It can be tough going at times, though.