Fix GPU copy kernels in periodicity handling

The option cupy_direct_copy=False in LBMPeriodicityHandling, generating dedicated copying kernels, was broken: it's implementation was trailing behind pystencils API changes by several versions. This MR fixes that, and also extends the tests for that feature.

Fixing this, bugs in pystencils relating to iteration slices in GPU kernels were revealed, see pystencils!422 (merged) and pystencils#103.

Depends on pystencils!422 (merged)

Closes #45 (closed)