use SolvOpt for attenuation coefficients instead of the old way

[komatits]

The new way to get the relaxation times for attenuation, which is described in http://komatitsch.free.fr/preprints/GJI_Lombard_2016.pdf , is called SolvOpt and used in SPECFEM2D, but not yet cut and pasted in SPECFEM3D and SPECFEM3D_GLOBE.

(cutting and pasting the call to the routine that computes tau_epsilon and tau_sigma).

[komatits]

Robin Lee robin DOT lee AT pg DOT canterbury DOT ac DOT nz is working on it and is going to implement it.

[komatits]

No easy way to do the full implementation because the code assumes the same tau_sigmas() (equally spaced in log of frequency) for Q_Kappa and Q_mu, while the general SolvOpt() implementation would require different ones.

[komatits]

However, remark from Alexis Bottero @bottero : calling SolvOpt() is much more expensive because it uses an involved nonlinear optimization process. Thus, if we do it permanently, let us probably turn it OFF by default.
cc Paul @pcristini

[komatits]

Not so sure it is a good idea, because very expensive if called by every GLL grid point. If called by a given region of the model only, once for the whole region, then very useful and with negligible cost.

[komatits]

Etienne @EtienneBachmann will see if it is worth doing that.

[komatits]

Not really needed.