Fix the assembly performance

[adtzlr]

due to non-contiguous region attributes.

import felupe as fem

mesh = fem.Cube(n=21)
region = fem.RegionHexahedron(mesh)

region.dhdX.flags
Out[1]: 
  C_CONTIGUOUS : True
  F_CONTIGUOUS : False
  OWNDATA : True
  WRITEABLE : True
  ALIGNED : True
  WRITEBACKIFCOPY : False

closes #827