However, can someone explain to me in simple terms why we would need in this context a large order of G and how it will contribute in making g^ab more secure
Background (which you likely already know), with El Gamal, the ciphertext is the pair $g^r, h^r \cdot m$ (where $h$ is the public key and $r$ is a random value); if we could figure out what the value $r$ is, we could recover $m$ (because we assume we know the value $h$).
So, one of the things that we need to make sure is that, given the value $g^r$, we can't deduce what $r$ is; this is known as the "discrete logarithm problem".
With that in mind, here are some of the math behind the scene:
We define the "order of G" to be the smallest value $k > 0$ for which $G^k = 1$. This is interesting because $G^r$ can take on only $k$ different values, $G^1, G^2, ..., G^{k-1}, G^k = 1$. If we keep on going, we'll end up repeating, starting back at $G^1$, and so that doesn't help us any.
If there are only $k$ different values of $r$ that give us different values of $G^r$, then if $k$ is small, what the attacker could do is just try all $k$ different values of $r$ and see if one works; if he can do that, he can recover the correct value of $r$ [1]. Actually, it turns out that if the attacker uses just a give of cleverness, he can perform this search with about $\sqrt{k}$ multiplications; hence if we want to ensure that the attacker must perform $2^{128}$ operations to attack the system (which is the current standard for "that is beyond anyone's capability"), then we need a $k$ at least $2^{256}$.
And, it turns out that there's another observation to be made: if $k$ is composite and has a prime factor $s$, the attacker can compute $r \bmod s$ with $\sqrt{s}$ operations (by using the same sort of cleverness); this reduces the strength of such a $G$.
using p = 2q+1 denotes that order of G is is 2 and q and that "g is then sometimes chosen to generate the order q subgroup of G, rather than G, so that the Legendre symbol of g^a never reveals the low order bit of a.
This talks about a common method of dealing with the above attacks (not the only method, mind).
It turns out that if $p = 2q+1$ prime, where $q$ is prime as well, and if $p \mod 8 = 7$ (something that Wikipedia forgot to mention), then the value $g=2$ always has order $q$; that is a large prime. What's so special about 2? Well, it makes computing the exponentiation a bit easier (as multiplying by 2 modulo p is quite cheap).
And, when it talks about the Legendre symbol revealing something about $g^a$, well, that is a special purpose method of finding $a \bmod 2$; in essence, it is the same approach as I referenced above to recover $a \bmod s$ for $s=2$; it works only if the order of $g$ has 2 as a factor. Because we picked a $g$ that has an odd order $q$, it doesn't work.
[1]: You might ask: even if $G^r$ is the same, wouldn't it be possible that $H^r$ take on different values? It turns out that no, that can't happen - if we finds any value $r$ that gives him the observed value of $G^r$, he can use that to decrypt.
