Fix LHA VFNS SV benchmark 2

[felixhekhorn]

Closes #215 - alternative implementation to #218

This PR is restoring the v0.10 benchmark (as #218 does), but avoids any division by $\xi_F$. The latter applies explicitly to the division of the masses on input (as done in #218) and that we don't feel comfortable with, and also to the former sv_exponentiated_shift (that is deleted in this PR). There are, of course, in exchange, some multiplications with $\xi_F$ appearing (as suggested by @andreab1997 ).

1. Note that I'm not doing the exact same in scheme A and B: 1) for scheme B I have to shift the last operator (as is our current understanding, that we only need to shift this one - though I'm no longer 100% sure about this) (Note that the implementation of this rule shifted from grid to ev_op) and 2) when doing this last step in scheme B the second and third argument to a1 are the same again (so unlike scheme A where the two arguments are always different)
2. We may want to discuss where to change the reference scale of $\alpha_s$ (that is needed to get the benchmark). Currently it's in the runner, but I wonder whether this is the best place ...
3. Also note that, as briefly discussed with @andreab1997, I realized (but was also true before) that on a strict mathematical basis scheme C EKOs are non-trivial (i.e. they do not coincide with the central theory), because the adjusted $\alpha_s$ definition also applies to them. To be even more precise: they should not affected by point 1, but by point 2

[andreab1997]

3. as briefly discussed with @Andrea1997, I realized (but was also true

I believe you tagged a random person 😂

[andreab1997]

Closes #215 - alternative implementation to #218

This PR is restoring the v0.10 benchmark (as #218 does), but avoids any division by ξF. The latter applies explicitly to the division of the masses on input (as done in #218) and that we don't feel comfortable with, and also to the former sv_exponentiated_shift (that is deleted in this PR). There are, of course, in exchange, some multiplications with ξF appearing (as suggested by @andreab1997 ).

1. Note that I'm not doing the exact same in scheme A and B: 1) for scheme B I have to shift the last operator (as is our current understanding, that we only need to shift this one - though I'm no longer 100% sure about this) (Note that the implementation of this rule shifted from grid to ev_op) and 2) when doing this last step in scheme B the second and third argument to a1 are the same again (so unlike scheme A where the two arguments are always different)
2. We may want to discuss where to change the reference scale of αs (that is needed to get the benchmark). Currently it's in the runner, but I wonder whether this is the best place ...
3. Also note that, as briefly discussed with @Andrea1997, I realized (but was also true before) that on a strict mathematical basis scheme C EKOs are non-trivial (i.e. they do not coincide with the central theory), because the adjusted αs definition also applies to them. To be even more precise: they should not affected by point 1, but by point 2

However, as you know, I agree with everything but I am still not sure about the differences between scheme B and A. Maybe we can discuss tomorrow?

[felixhekhorn]

Close in favor of #222