Get the full LJ optimization running on Travis

[ctk3b]

Once HOOMD becomes conda installable, this will hopefully be much easier. We may also need to reduce the system size in order to get enough sampling within the time allotted for a Travis build on a single CPU. Even without that, however, we should be able to do more complete testing.

[tcmoore3]

We may also need to reduce the system size in order to get enough sampling within the time allotted for a Travis build on a single CPU...

Depends on how things scale. Could just dump the trajectory more often