npy reader and loader should have the same behaviour as in numpy

[sannant]

How the numpy reader works:

• read metadata (shaoem, dtype and whether fortran_order)
• load data into a 1d array arr (is at the same time C and fortran contiguous)
• if fortran_order ==False arr.shape=shape (at this moment the array becomes C_contiguous)
• if fortran_order ==True: arr.shape=shape[::-1]; arr = arr.transpose() (transpose toggles C_contiguous to F_contiguous and swaps the shape back again)