revisit default LD coefficients for training data

[SteveOv]

John has shown me some new material he's produced the his dEBs module where he's plotted the behaviour of the quad & pow2 coefficients for various bands when varying the $T_{eff}$ while keeping $\log{g}=5$, [M/H]=0 and $V_{micro}=2$. We should be able to refine our general purpose coefficients for generating the training data from similar plots.