Yes, they do. Something complicated about quantum field theory is that we often hide the countless tensor products that are in effect. If we use $a,b$ as spinor indices (i.e., indices labeling the spinorial entries of spinorial matrices and spinors), we can write
\begin{align}
(\gamma^\mu \displaystyle{\not}a \gamma_\mu)_{ab} &= \gamma_{ac}^\mu \displaystyle{\not}a_{cd} \gamma_{\mu,db}, \\
&= \gamma_{ac}^\mu a_{\nu} \gamma^{\nu}_{cd} \gamma_{\mu,db}.
\end{align}
This time, all terms are numbers, since we are working in components. Thus,
\begin{align}
(\gamma^\mu \displaystyle{\not}a \gamma_\mu)_{ab} &= \gamma_{ac}^\mu a_{\nu} \gamma^{\nu}_{cd} \gamma_{\mu,db}, \\
&= a_{\nu} \gamma_{ac}^\mu \gamma^{\nu}_{cd} \gamma_{\mu,db}, \\
&= a_{\nu} (\gamma^\mu \gamma^{\nu} \gamma_{\mu})_{ab},
\end{align}
at which point you can just follow your calculation and things run smoothly.
The key point is noticing that four-vector do not carry spinor indices.