allow for choosing shear virial weight

[brucefan1983]

In most cases, the shear virials/pressures in the training data are much smaller than the diagonal ones and it is might be beneficial to increase the weight for the shear part. Now one can use

lambda_shear <value>

in nep.in to set such a weight. Someone tested that a value of 5 can make the prediction of C_44 much more accurate in some material.

To solve #295

[erhart1]

Looks good to me.
I would probably rather include more specifically deformed configurations but this seems works for some.
Do you still want to add a page to the documentation?

[brucefan1983]

Yes, I will update doc.

[brucefan1983]

Looks good to me. I would probably rather include more specifically deformed configurations but this seems works for some. Do you still want to add a page to the documentation?

In some cases it is even insufficient to include specifically deformed configurations according to tests by Zhiqiang (I think he has contact you for entory maximization). The problem is the shear virials/moduli are much smaller than diagonal ones in some materials (one order of magnitude smaller).

[brucefan1983]

I messed up by doing something accidently in master...
perhaps need to start this PR over

[brucefan1983]

I fixed the conflict and it is ready to merge to master.

documentation has been directly committed to master.