perf: optimize Verify and BatchVerifyMultiPoints methods

[yelhousni]

This PR ports two optimizations from gnark-crypto:

• Re-arranging the Verify pairing equation so that instead of doing a G1 scalar mul and a G2 scalar mul, we do a G1 2-MSM with Strauss-Shamir trick.

// [f(z)]G₁ + [-z]([H(α)]G₁) = [f(z) - z*H(α)]G₁

• Re-arranging the pairing equations in Verify and BatchVerifyMultiPoints makes it possible to pre-compute the lines in pairing computation corresponding to G₂ and [α]G₂

 // e([f(α)-f(z)+aH(α)]G₁], G₂).e([-H(α)]G₁, [α]G₂) == 1

These pre-computed lines can be stored in the SRS.

Bench

On a z1d.large AWS machine

goos: linux
goarch: amd64
pkg: github.com/crate-crypto/go-kzg-4844/internal/kzg
cpu: Intel(R) Xeon(R) Platinum 8151 CPU @ 3.40GHz

First optimisation saving for Verify:

benchmark                old ns/op     new ns/op     delta
BenchmarkKZGVerify-2     1121344       982636        -12.37%

The two optimisations combined:

benchmark                old ns/op     new ns/op     delta
BenchmarkKZGVerify-2     1121344       822792        -26.62%

[yelhousni]

Last commit reverts the use of pre-computed lines which are not currently compatible with Ethereum SRS.