We have the sum
$$S_{n,q} = \sum_{k=q}^{\lfloor (n-1)/2\rfloor}
{k\choose q} {n\choose 2k+1}.$$
Introducing
$${n\choose 2k+1} = {n\choose n-2k-1} =
\frac{1}{2\pi i}
\int_{|z|=\epsilon}
\frac{1}{z^{n-2k}} (1+z)^n
\; dz$$
this controls the range and we may extend $k$ to infinity, getting for
the sum
$$\frac{1}{2\pi i}
\int_{|z|=\epsilon}
\frac{1}{z^{n}} (1+z)^n
\sum_{k\ge q} {k\choose q} z^{2k}
\; dz
\\ = \frac{1}{2\pi i}
\int_{|z|=\epsilon}
\frac{1}{z^{n-2q}} (1+z)^n
\sum_{k\ge 0} {k+q\choose q} z^{2k}
\; dz
\\ = \frac{1}{2\pi i}
\int_{|z|=\epsilon}
\frac{1}{z^{n-2q}} (1+z)^n
\frac{1}{(1-z^2)^{q+1}}
\; dz
\\ = \frac{1}{2\pi i}
\int_{|z|=\epsilon}
\frac{1}{z^{n-2q}} (1+z)^{n-q-1}
\frac{1}{(1-z)^{q+1}}
\; dz.$$
This is
$$\sum_{p=0}^{n-2q-1} {n-q-1\choose p} {n-2q-1-p+q\choose q}
\\ = \sum_{p=0}^{n-2q-1} {n-q-1\choose n-q-p-1} {n-q-1-p\choose q}.$$
Observe that
$${n-q-1\choose n-q-p-1} {n-q-1-p\choose q}
= \frac{(n-q-1)!}{p! q! (n-2q-1-p)!}
\\ = {n-q-1\choose q} {n-2q-1\choose p}$$
so that the sum becomes
$${n-q-1\choose q} \sum_{p=0}^{n-2q-1} {n-2q-1\choose p}
= 2^{n-2q-1} {n-q-1\choose q}.$$
where we have $q\le \lfloor (n-1)/2\rfloor.$ If this is to be a
multiple of $2^{n-2q}$ then the binomial coefficient must be even. We
apply Lucas'
Theorem which says
that a binomial coefficient ${n\choose p}$ is odd iff all the bits of
$p$ are less than or equal to the corresponding bits of $n.$ Suppose
first that $n$ is not a power of two and take $q = n - 2^{\lfloor
\log_2 n\rfloor}.$ We have $q\ge 1$ as required. To verify the upper
range start from $\lceil (n-1)/2\rceil \le n/2$ which follows by
cases. This is $\lceil (n-1)/2\rceil \le 2^{-1 + \log_2 n}$ which is
less than $2^{\lfloor \log_2 n\rfloor}$ so that $n-1 - \lfloor
(n-1)/2\rfloor \lt 2^{\lfloor \log_2 n\rfloor}$ or $n - 2^{\lfloor
\log_2 n\rfloor} \lt \lfloor (n-1)/2\rfloor + 1$ or $n - 2^{\lfloor
\log_2 n\rfloor} \le \lfloor (n-1)/2\rfloor.$ This proves that the
choice of $q$ is within the range. We thus have $n-q-1 = 2^{\lfloor
\log_2 n\rfloor} - 1$, a string of one bits, which are larger than or
equal to the bits of $q$ by inspection. Hence we have found an odd
binomial coefficient and the claim does not hold when $n$ is not a
power of two. The remaining case is for $n$ a power of two. We thus
have $n-1$ a string of one bits. Hence the bits corresponding to $q$
in $n-1-q$ are the bits of $q$, but flipped. Now since all $q$ have at
least one bit that is set we get a correspondong zero bit in $n-1-q.$
Hence the binomial coefficient is even. This works for all $q.$
Therefore the answer to the query is that $S_{n,q}$ is divisible by
$2^{n-2q}$ with $q$ in the proposed range iff $n$ is a power of two.
Some of this material is sourced from the following MSE
link.