The decay $\pi^+ \to \mu^+\nu_\mu$ is a bit of a "centaur": The pion couples to the hadronic current and disappears into the vacuum, a strong interaction; while the hadronic current weakly converts to a $W^+$ which then weakly converts to the antimuon-neutrino pair.
You are done with (perturbative) Feynman diagrams, here: the W has been integrated out to produce the 4-Fermi contact interaction (28.31), as described in your text, leaving only $G_F$ as its only trace! You are really asking about notation of a plugin. This is the language of "current algebra", or "current-current interaction", framing the field before the advent of the standard model, and ingeniously bypassing our ignorance about the nonperturbative chiral symmetry breaking, connecting quarks to hadrons, (pseudoscalars here), a subtle process.
As your text indicates, (28.30) leads to the "horse half" $J_\mu^{5a}=F_\pi\partial_\mu \pi^a$, the low-energy hadronic equivalent of (28.20), the high-energy "human" half--the quark axial current $J_\mu^{5a}=\bar q \tau^a\gamma_\mu \gamma^5 q$ fitting into the L weak currents of (28.31),
$$F_\pi\partial_\mu \pi^a \longleftrightarrow \bar q \tau^a\gamma_\mu \gamma^5 q~~. $$
So, the only pieces of (28.32) contributing to the decay $\pi^+ \to \mu^+\nu_\mu$ summarizing all effective low-energy weak interactions in (28.31) is, really,
$$
J_\mu^{L+}
\approx
- \frac{F_\pi}{\sqrt 2}\partial_\mu \pi^+ + \overline{\psi_{\mu}}\gamma^\mu (1 - \gamma^5) \psi_{\nu_\mu}
+
\cdots \tag{28.32'}
$$
where, like Matt, I have relegated all useless pieces to the ellipses (...).$^\natural$ The first term kills a $\pi^+$ and the second kills a $\mu^+$ and a muon neutrino. The h.c. $J_\mu^{L-}$ multiplying it creates these states, so the cross term destroys a
$\pi^+$ and creates a $\mu^+$ and a neutrino.
Taking matrix elements $\langle \mu^+ \nu_\mu | \frac{G_F}{\sqrt 2} J^{L+}\cdot J^{L-} | \pi^+\rangle$ yields
$$
\mathcal{M}
(\pi^+ \to \mu^+\nu_\mu)
=
\frac{G_F}{\sqrt{2}}
F_\pi p_\mu \langle \mu^+ \nu_\mu | \bar{\psi}_{\nu_{\mu}}
\gamma^\mu (1- \gamma^5)\psi_\mu|0\rangle. \tag{28.33}
$$
$^\natural$ I seem to disagree with Matt's perverse (28.32) with the L projector in the quark instead of the leptonic piece, ... puzzling, probably designed to keep signs straight... (28.33) equates an amplitude to an operator?? They might be fixed like lots of typos in subsequent editions. Worry about the concepts involved, not the accuracy of his normalizations.