Accuracy in Compton Spherical Layout

[szabo137]

Here, we could have some imprecisions:

QEDprocesses.jl/src/processes/one_photon_compton/perturbative/kinematics.jl

Lines 34 to 36 in e47ef46

	Et = getE(pt)
	rho2_t = getMag2(pt)
	s = Et^2 - rho2_t

[AntonReinhard]

My current best attempt is the following:

@inline function _pert_Compton_omega_prime(pt, cth)
    e = getE(pt)
    x = getX(pt)
    y = getY(pt)
    z = getZ(pt)

    s = (e - x - y - z) * (e + x + y + z) + 2(x * y + x * z + y * z) - 1

    return s / (2 * (e - getMag(pt) * cth))
end

But (e - getMag(pt) * cth) can still get very small when e ≈ getMag(pt) * cth, which leads to catastrophic cancellation, and I haven't found a way to avoid that yet.