In key generation, figure out if we need l_0(X) computed more efficiently

[ed255]

In the key generation we find the following TODO:

halo2/halo2_proofs/src/plonk/keygen.rs

Lines 360 to 365 in 73408a1

	// Compute l_0(X)
	// TODO: this can be done more efficiently
	let mut l0 = vk.domain.empty_lagrange();
	l0[0] = C::Scalar::ONE;
	let l0 = vk.domain.lagrange_to_coeff(l0);
	let l0 = vk.domain.coeff_to_extended(l0);

We need to figure out:

1. Is it worth doing this part more efficiently? In my opinion, if this part represents a <1% of the key gen time, there's no need to improve it.
2. If we want to do it more efficiently, figure out how, and do it.