enforced double precision calculations in *Group.center()

[zemanj]

Fixes #1930 (at least in part)

Changes made in this Pull Request:

• enforced calculations in *Group.center() to be performed in double precision
• for compatibility reasons, added automatic type conversion to dtype=np.float32 for input coordinate arrays of lib.distances.apply_PBC(), lib.distances.distance_array(), and lib.distances.self_distance_array()
• arrays returned by *Group.center() will be of dtype=np.float64 unless (compound != 'group' and pbc==True). This exception could be eliminated by providing a double-precision version of lib.distances.apply_PBC().

PR Checklist

• Tests?
• Docs?
• CHANGELOG updated?
• Issue raised/referenced?

[richardjgowers]

@zemanj if you're doing a few you may as well do all the functions in distances (eg calc_bonds)

[coveralls]

Coverage increased (+0.008%) to 89.801% when pulling de3deb7 on zemanj:group_center into 70b9e9f on MDAnalysis:develop.

[zemanj]

@richardjgowers I only did the ones that were needed to pass the tests since I didn't know whether you guys are ok with what I'm doing. Sure I can do that for the remaining functions as well.
Updates for tests, docs, and changelog are also needed.

[orbeckst]

@richardjgowers did you already take this PR into account in your PR #1939?

If not, should we merge this one "as is" and then worry about double precision in your PR (rebased) or do a new one to add remaining functionality?

[richardjgowers]

needs typecasting on all accessible functions in lib.distances

[richardjgowers]

@orbeckst my PR just pushes single angle/etc calculations into the existing functions. This PR makes those existing functions work with double precision arguments. So they're complimentary and non conflicting

[zemanj]

Recent changes:

• Added typecasting of input coordinate arrays to np.float32 for all accessible functions in lib.distances
• Changed docstrings accordingly
• Removed respective "wrong dtype" tests, parametrized tests for np.float32 and np.float64 instead.

Open questions:

• Do we need tests for other dtypes?

Will add CHANGELOG entry once this passes.

[richardjgowers]

@zemanj I think float32 & 64 is enough

[richardjgowers]

looks awesome!