misc: handle connection errors even before the timeout

[patrickhulce]

Summary
Should fix travis and give us better error messages when Chrome crashes unexpectedly.

Previously we only handled connection errors after we reached our timeout when we should be handling them as soon as we try to connect in case Chrome crashes unexpectedly.

Related Issues/PRs
#7534

[patrickhulce]

So the errors smoke test has been fixed :)

but the NO_TRACING_STARTED is still very in our face, perhaps #7122 is more of a problem than we realized :(

[brendankenny]

so what changed with 74 exactly?

[patrickhulce]

so what changed with 74 exactly?

Chrome's shutdown procedure is clearly different though I'm not sure what exactly. If I had to guess I would say it's related to network service since that's dramatically changed how Chrome shuts down (usually takes 4-6s now locally).

EDIT: If you were asking what the failure mode was before this, it was that Chrome closes the connection before get a chance to close it ourselves and because we don't handle the error event on the request, node exits prematurely before we can log anything about the error.