ENH: improvements for nastran surface writer (#1571)

- avoid face copying.
  Maintain separate offsets/list for non tri/quad face decomposition,
  which eliminates copying for tri/quad types that represent the bulk
  of geometries

- report inappropriate use of PLOAD2 for higher-ranks only once per
  field instead of per face.  For this case, write its magnitude
  instead of 0.

- perform field output scaling prior to calling the write face
  function. This will make it easier to handle different per-field
  scaling in the future (#1612)

BUG: nastran quad written as "CTRIA3" instead of "CQUAD4"