working.

latest pypy release

called from .travis.yml and stop looking at coverage on pypy, since performance
is horrific (4x longer when using coverage).

- Turn on timing for tests to aid in profiling
- Make sure we install coverage and nose-timer packages
- Turn segid test on in general
- Rename several tests
- Fix failure caused when OpenMM is not available

the OS X build into a minimal build with as few packages as possible (this
should save ~10 minutes on the aggregate build time, and we were within 1 minute
of going over the limit, I believe).

differentiated between two different residues in a CHARMM PDB file. So now
ParmEd matches the number of residues for overflow.pdb that I determine by hand
(the old code had combined two of the single-residue carbohydrate residues).

This also converts all of the OpenMM PME tests back to using the Reference
platform so that I can go back to using the openmm conda package.  The
openmm-dev conda package is too unreliable to count on, and openmm 6.3 hangs
indefinitely on the CpuPme plugin.

This fixes a couple subtle bugs from missing imports and a bunch of other minor
things tagged by pyflakes. It also adds a pyflakes check to the Travis CI script
that will fail if any issues are flagged. However, pyflakes flags a lot of
issues that are not really issues, and are IMO the best way to do something. So
there is a wrapper script around pyflakes that filters out these "OK" failures
(including any failure with 3rd-party libraries that I don't want to modify
heavily).

Convert back to using the CPU platform for PME and pull in the openmm-dev
package from conda. See if the recent change to the CPU PME plugin fixes the
weird hangs that I was seeing with Travis *all the time*.

setup

it locally

Should give faster builds and more memory (!!)

This makes a number of changes:

- numpy is now a hard requirement
- coordinates always has shape (frame, natom, 3)
- you can set coordinates, and it is reshaped to a (frame, natom, 3) ndarray,
  and the first frame's coordinates are assigned to the xx, xy, and xz
  attributes of each atom in the Structure
- did the same thing with the 'box' attribute (which is also always a
  numpy.ndarray now)
- add tests for this new functionality