Pre-release smoke test on travis doesn't work correctly

[paulkaplan]

Not sure how I didn't realize this before, but the "old browser" smoke test that gets run when PRs are made against master doesn't really work correctly, since it is doing it against the gh-pages build but that hasn't been deployed at the time when the test is run. In order to really work right now you have to rebuild on travis a second time after the initial build is uploaded to gh-pages. I'm silly :(

@rschamp I know you were probably trying to tell me this at some point and I just wasn't understanding. I'm not sure an additional travis stage would work because of the caching on gh-pages (10mins?) so we might have to use the local server + tunneling to sauce that we planned to move to eventually. Until then I'll just run those tests manually and/or re-trigger the build twice on the smoke => master PR.

/cc @BryceLTaylor maybe we could pair on setting up the sauce tunnel, since you've been doing similar work for www right?

[BryceLTaylor]

@paulkaplan I've been working with Saucelabs for www, but haven't set up a tunnel, so I'm not that familiar with that aspect yet.

[rschamp]

I think testing on the local build will end up being the most reliable and useful for running these smoke tests. Using a tunnel would let us run these tests on every build, so we could know which PR actually broke them before it got merged. I have set up local tunnels before, and am happy to work with both you and @BryceLTaylor on getting that set up in Travis (would be a useful thing to have on www as well).

I suppose we would want to know if the gh-pages build failed to occur (on GitHub's side), but that's not really the point of these tests.