optimize generate_box_mesh (#458, part 1)

* optimize generate_box_mesh

* optimize boundary processing loop

* fix type annotation

* parameterize dtypes

* simplify definition of a0/a1 for 1D

* use np.column_stack() instead of np.array([...]).T

* avoid potentially creating temporary arrays in vertex assignment

* *_crit -> *_face

Author: majosm