There was a nasty bug in io_orientations in nibabel, that could cause
the process to hang when getting orientations for affines will all-zero
columns. To allow use of nibabel 1.2.0, copy the fix into our fixes
tree until we depend on a later nibabel.
I was trying to be too fancy, because I was trying to return a valid xyz
affine for coordmaps that did not have output order X, Y, Z. In the
process I got confused and failed to check the correspondence of input
and output axes, letting past cases where input axes had been dropped or
were in the wrong order.