Make caching behavior more like Selenium

[jferris]

We disabled caching entirely, because it sometimes led to unexpected
behavior in tests, and Qt 4.8 didn't behave the same as Qt 5.x.

However, this introduced issues:

• Selenium does perform caching, so capybara-webkit behaved differently
  when switching drivers
• Without any caching, assets were sometimes requested twice for the
  same page

This commit re-enables caching and attempts to improve the way we behave
on cached pages so that testing is less unpredictable. It adds tests to
ensure we continue to behave similarly to Selenium.

[mhoran]

It looks like one of the builds failed here due to an ordering problem.

[jferris]

Just pushed 4748bb5 to hopefully address this.

[mhoran]

This seems reasonable to me, so long as we're bug for bug compatible with Selenium!

Now that I am thinking about this again, I recall that Rails incorrectly returned 304 under certain circumstances in an app that exhibited this cache issue. We only saw the bug under Qt 5, so there's certainly something that changed there but I'm not sure what. Adding a random query-string to the page resolved the issue and returned the correct content.

Given Selenium would likely have behaved the same way, I'd say this is the right way to go. My only remaining concern is the behavior of the headers, body and contentType properties of cached pages.