LDAU+SOC, taking into account local spin orientation

LDA+U as implemented now does not work properly with SOC. Besides more sophisticated corrections which might be needed (??, I ignore this at the moment), there is an obvious problem related with the formulation of LDA+U implemented at the moment. This formulation only uses "up" and "down" spin populations with respect to an arbitrary quantization axis (z-axis) and ignores local spin orientation. This can be corrected very easily by a local rotation, pretty much as it is done before computing LDA xc energy and potential (that requires the input of the density and local magnetization).