Is it always true that $p\geq 3$ not a prime divisor of \begin{equation*} \sum_{t=0}^{\frac{p-1}{m}-1}\sum_{s=0}^{\frac{p-1}{m}-1}\binom{tm}{sm} \end{equation*} where $m$ divides $p-1~?$
The following is my attempt, and I made some simplifications in order to conclude something. But it seems not to be very helpful. Any comments or advice would be grateful.
\begin{align*} \sum_{t=0}^{\frac{p-1}{m}-1}\sum_{s=0}^{\frac{p-1}{m}-1}\binom{tm}{sm}&=\sum_{t=0}^{\frac{p-1}{m}-1}\left(\binom{tm}{0}+\binom{tm}{m}+\binom{tm}{2m}+\cdots+\binom{tm}{p-1-m}\right)\\ &=\sum_{t=0}^{\frac{p-1}{m}-1}\left(1+\binom{tm}{m}+\binom{tm}{2m}+\cdots+\binom{tm}{p-1-m}\right)\\ &=\frac{p-1}{m}+\sum_{t=0}^{\frac{p-1}{m}-1}\binom{tm}{m}+\sum_{t=0}^{\frac{p-1}{m}-1}\binom{tm}{2m}+\cdots+\sum_{t=0}^{\frac{p-1}{m}-1}\binom{tm}{p-1-m}\\ &=\frac{p-1}{m}+\sum_{t=1}^{\frac{p-1}{m}-1}\binom{tm}{m}+\sum_{t=2}^{\frac{p-1}{m}-1}\binom{tm}{2m}+\cdots+\sum_{t=\frac{p-1}{m}-1}^{\frac{p-1}{m}-1}\binom{tm}{p-1-m}\\ &=1+\frac{p-1}{m}+\sum_{t=1}^{\frac{p-1}{m}-1}\binom{tm}{m}+\sum_{t=2}^{\frac{p-1}{m}-1}\binom{tm}{2m}+\cdots+\sum_{t=\frac{p-1}{m}-2}^{\frac{p-1}{m}-1}\binom{tm}{p-1-2m} \end{align*}
$\def\ZZ{\mathbb{Z}}\def\FF{\mathbb{F}}\def\CC{\mathbb{C}}$I will prove the following:
Thanks to mathlove's answer for many of the key ideas in this answer.
To start with, take no hypotheses on $m$ and let $p$ be a prime which is $1 \bmod m$. Since $p \equiv 1 \bmod m$, there are $m$ distinct $m$-th roots of unity in $\FF_p$, call them $\gamma_1$, $\gamma_2$, ..., $\gamma_m$.
Note that we have $$\sum_{n=0}^{p-2} (\gamma_i+\gamma_j)^n = \sum_{n=0}^{p-2} \sum_{k=0}^n \binom{n}{k} \gamma_i^k \gamma_j^{n-k}.$$ Summing on $i$ and $j$, we get $$\sum_{i,j=1}^m \sum_{n=0}^{p-2} (\gamma_i+\gamma_j)^n = \sum_{n=0}^{p-2} \sum_{k=0}^n \binom{n}{k} \sum_{i,j=1}^m \gamma_i^k \gamma_j^{n-k}.$$ The inner sum is $m^2$ if $k \equiv n-k \equiv 0 \bmod m$, and is $0$ otherwise, so we get that $\sum_{i,j=1}^m \sum_{n=0}^{p-2} (\gamma_i+\gamma_j)^n$ is $m^2$ times the sum in question.
Now, for $a \in \FF_p$, we have $$\sum_{n=0}^{p-2} a^n = \begin{cases} 1 & a =0 \\ p-1 & a = 1 \\ 0 & \text{otherwise} \end{cases}.$$ To see the last case, note that, for $a \neq 0$, $1$, we have $$\sum_{n=0}^{p-2} a^n = \frac{a^{p-1}-1}{a-1} = 0.$$
Thus, we need to understand whether $\gamma_i+\gamma_j$ could be $0$ or $1$. Our result will follow from proving the following two lemmas:
Lemma 1 is easier.
Proof of Lemma 1 Suppose that $\gamma_i+\gamma_j = 0$. Then $-1 = \gamma_i \gamma_j^{-1}$, so $-1$ is an $m$-th root of unity. But $m$ is odd and $p>2$, a contradiction. $\square$
To prove Lemma 2, we use resultants.
Lemma 3 For $p$ a prime which is $1 \bmod m$, there are two $m$-th roots of unity in $\FF_p$ with sum $1$ if and only if $p$ divides one of the $m^2$ many resultants $R(x^a+x^b-1, x^m-1)$ for $0 \leq a,b \leq m-1$.
Proof: We recall that $R(x^a+x^b-1, x^m-1) = 0 \bmod p$ if and only if $x^a+x^b-1$ and $x^m-1$ have a common root in the algebraic closure $\overline{\FF}_p$. Since $p \equiv 1 \bmod m$, all roots of $x^m-1$ in $\overline{\FF}_p$ in fact lie in $\FF_p$.
So, if $\zeta$ is a common root of $x^a+x^b-1$ and $x^m-1$, then $\zeta^a$ and $\zeta^b$ are two $m$-th roots of unity in $\FF_p$ summing to $1$.
Conversely, suppose that we have two $m$-th roots of unity, $\alpha$ and $\beta$, in $\FF_p$, which sum to $1$. The group of $m$-th roots of unity in $\FF_p$ is cyclic; let $\zeta$ be a generator and let $\alpha = \zeta^a$ and $\beta = \zeta^b$. Then $\zeta$ is a common root of $x^m-1$ and $x^a+x^b-1$ in $\FF_p$. So we have proven the equivalence. $\square$
Lemma 1 will follow from Lemma 3 as long as none of the resultants $R(x^a+x^b-1, x^m-1)$ are $0$ (just take $N$ larger than any of these resultants). So, to finish the problem, we must show:
Lemma 4: Let $m$ be not divisible by $6$. Then the restultant $R(x^a+x^b-1, x^m-1)$ is not $0$ for any $0 \leq a,b \leq m-1$.
Proof: If $R(x^a+x^b-1, x^m-1)=0$, then the polynomials $x^a+x^b-1$ and $x^m-1$ have a common root in $\CC$. So there are two $m$-th roots of unity in $\CC$ which add to $1$. But the only two numbers on the unit circle which sum to $1$ are $\tfrac{1 \pm \sqrt{-3}}{2}$, which are primitive $6$-th roots of unity. So, since $m$ is not divisible by $6$, these are not $m$-th roots of unity. $\square$
This concludes the proof of the Theorem.
I'll note that the prime factors of these resultants can be surprisingly large! Here is a table of all the factors for the first few values of $m$:
I haven't checked this as carefully, but you should be able to use the same argument to work out what the sum equals when $m$ is even, and $p$ is large enough not to divide any of the nonzero resultants. I believe that, for $m \equiv 2,4 \bmod 6$, the sum is $\tfrac{1}{m^2} m$ (coming from the $m$ pairs $(\zeta, - \zeta)$ of $m$-th roots of unity summing to $0$) and, for $m \equiv 0 \bmod 6$, we get $\tfrac{1}{m^2} (m-2)$, where we add in the $m$ pairs $(\zeta, - \zeta)$ as above, and then subtract of $2$ for $(\eta, \eta^{-1})$ and $(\eta^{-1}, \eta)$, where $\eta$ is a primitive $6$-th root of unity.