Inconsistent k-point grids in QE

[ondracka]

Just something I noticed in the recent QE work (but no time to fix now, so making a notice for later).

The k-grid in method and eigenvalues is inconsistent (most likely the kgrid in method is wrong.) Specifically the k_mesh in method section is using Cartesian coordinates while the k_mesh in eigenvalues is using crystal coordinates.

QE prints kpoints at several places:
kpoint in Cartesian coordinates (in units of 2pi/alat) are printed in the header
kpoint in crystal coordinates could be printed in the header as well (probably when verbosity is set to high), see for example the Pt test case
kpoints in Cartesian coordinates are printed again with the corresponding eigenvalues.

Right now both method.k_mesh.points section seems to use the values from eigenvalues section but they are transformed to the crystal coordinates only for the kpoints saved in the eigenvalues.

We should probably:

• If the kpoints in crystal coordinates are printed in the header than use them everywhere
• If the crystal coordinates are not there, we should still use the cartesian coordinates from the header (there is more precision than in the eigenvalues section) do the conversion to crystal coordinates once and reuse this both for the eigenvalues and k_mesh section.