Fix CircleCI Build

[krotscheck]

This patch attempts to fix the continually breaking circleci build. The changes made were:

• Switch to using circleci v2 syntax, as the v1 syntax will be deprecated in August. Also, it permits running the build locally using the circleci commandline tool.
• Adjust timeouts on error tests to fail consistently on high-performance machines.
• Correct error formatting on error tests.

Fixes #4810

[krotscheck]

Right, so, I have no idea why this works. I have two hypotheses:

1. Slightly newer version of node-6.
2. CircleCI bumped their container memory allowance to 4Gb with v2.

But, hey! Circle's green again, so 🍻 all around.

[krotscheck]

@qiyigg 🎁 for ya. Do you mind taking a look so all the queued-up PR's can rebase?

[qiyigg]

@krotscheck Cool! Thanks a lot. By the way, do you find the root cause of the previous failure?

[krotscheck]

@qiyigg - nothing other than supposition. What I observed is that failing builds locally tended to fail at the end of a spec, rather than at the beginning. There's a whole variety of bugs in the geckodriver project (mozilla/geckodriver#966) regarding the correct order of driver.close() and driver.quit(), and given that the chromedriver also seems to linger if jasmine exits with an error, I believe the driver teardown logic itself is the root cause. i.e. lots of driver processes get started, not cleaned up, and suddenly the container runs out of memory.

From what I saw in the teardown logic (driverProvider.ts), it seemed like the error case on getSession() isn't really handled at all, and we'd need to go through to explore the various edge cases to make sure everything is cleaned up properly.

[qiyigg]

ok, thanks, merged.