core/io: fix H5MD time/step datasets off-by-one vs value datasets

[RudolfWeeber]

create_dims() for rank-1 datasets (step/time) returned an initial shape
of {data_dim} (non-zero), while rank-2/3 value datasets started at {0,...}.
As a result id/step and id/time were seeded with one fill element at index 0,
and every subsequent write landed at offset extents[0] of the value dataset,
misaligning the time axis by one frame for the entire trajectory.

Fix: return {0ul} for rank-1 in create_dims(), matching the value
datasets (src/core/io/writer/h5md_core.cpp).

Regression test: test_time_step_alignment added to H5mdTests in
testsuite/python/h5md.py.

Fixes bug #1 of the adversarial core bug sweep.

[RudolfWeeber]

@hidekb what's your take on this?

[hidekb]

This suggestion is essentially correct. However, this modification causes an error in the test hdmd.py. I try to fix it.