Fix phantomjs request aborted edge case - #82

I experienced a difficult to reproduce edge case on an Ember application with > 500 tests:

When running with coverage on the v0.3.8 tag, my test server crashes right after the tests complete with the following error:

Error: request aborted
    at IncomingMessage.onAborted ([snip]/ember-cli-code-coverage/node_modules/raw-body/index.js:269:10)
    at emitNone (events.js:86:13)
    at IncomingMessage.emit (events.js:185:7)
    at abortIncoming (_http_server.js:283:11)
    at Socket.serverSocketCloseListener (_http_server.js:296:5)
    at emitOne (events.js:101:20)
    at Socket.emit (events.js:188:7)
    at TCP._handle.close [as _onclose] (net.js:498:12)

This wasn't happening on the version we were using before (v0.2.2), so I figured it was a change in the recent minor bumps that caused the breakage. I bisected tags, and reproduced the issue in v0.3.7 and not v0.3.5 (v0.3.6 doesn't work with our codebase at all due to a throw that was downgraded to a warn in v0.3.7).

Looking at the compare between those versions, I couldn't see anything that looked even the slightest bit innocuous, so I started a long and fruitless journey through ember-cli, testem, and friends to try to see if I could pin down a source anywhere.

I also set up a new ember application and prodded it for a good amount of time looking for a repro case to no avail. Unfortunately it's a consistent error in a private repo (every jenkins build since I bumped ember-cli-code-coverage has broken along with local builds), but I have no repro case to offer.

As an additional annoyance, I was unsuccessful (in a workday) in finding the real source of the issue in the testem flow, but can say with certainty that it is not caused by:

• a long timeout on the /write-coverage endpoint
• onload use rather than readystatechange events
• calling the sendCoverage callback in the same event loop as the coverage XHR onload event (I tried setting various timeouts before calling callback, same result)

The only difference I could find in my large suite vs a brand new one was the fact that (oddly), Testem was emitting 'all-test-results' multiple (3) times. I figured this was the issue (it resolves the first XHR with 3 in flight, and the second one is cancelled by a disconnect). However, I was unable to fix this by only allowing sendCoverage to be called once, leading me to believe that either my suite is spinning up multiple runners, or the issue is unrelated (or in an alternative listener).

Through this exploration, I found a few different ways to workaround the failure, the simplest of which being contained in this PR (just make the XHR sync when inside PhantomJS) .

I'm almost 100% certain the issue I'm experiencing is not caused by this repo, but this seems like the most reasonable place to add a workaround in lieu of fixing the root cause in a dependency lib, since the use case (sending an XHR from test-body-footer.html on test complete) is fairly esoteric and hunting down the root cause is likely going to take longer than I have available to work on this case.