Review the use of timeouts

[santigimeno]

I'm not sure if this deserves an issue or if there's anything actionable, but anyway...

Lately, most of the failing tests I've found: nodejs/node#5342, nodejs/node#5339, nodejs/node#5129 were flaky because of the use (misuse?) of timers. It seems to me that a lot of times, the use of timers is not needed, as we can rely on the test runner timeout, and they become a source of flakiness. I could not agree more with this post from @Trott nodejs/node#5342 (comment). Should we set a position on this and maybe document it somewhere?

[Trott]

I would love to get some consensus on whether arbitrary timers is (on the one hand) a valid way to provide a helpful error message to people running tests without the wrapper or (on the other hand) an anti-pattern that introduces maddening flakiness into our tests. (Spoiler alert: I think it's an anti-pattern.)

[santigimeno]

Closing in favor of #27