Don't attempt to warm the Capybara cache outside of CI

We initially started doing this because the first feature test to use the JavaScript driver would often time out due to asset compilation on CI. We should still do it in CI since the cost is (relatively) low, but doing so for every local run is both tiresome and generally unnecessary, as developer machines will be faster than most shared CI runners, and the cost of retrying a test after a timeout is much, much lower.