write_timestep() needs updates to generate a correct XDMF file when saving data from discontinuous mesh variables

[gthyagi]

Currently, write_timestep() generates XDMF files with vertex datasets. For discontinuous mesh variables, the XDMF file should include cell datasets instead. This will also require some modifications to checkpoint_xdmf().

[julesghub]

Do you know how petsc handles discontinous variables I/O? We ought to leverage that if appropriate.

[gthyagi]

check this #328