Promises and tests

[MajorBreakfast]

• Promises: Broccoli uses promises pretty extensively. Consequently they're the perfect match for quick-temp
• Tests: For the three functions to ensure they work as advertised

Note: This PR is a breaking change. There is no way not make it non-breaking because going async is inherently a breaking change.

[joliss]

Thanks for the PR! Hm, what's the motivation for using promises? Is there a use case that is not covered by the original synchronous implementation?

[MajorBreakfast]

Sync has problems. Isaac also recommends using async https://github.com/isaacs/rimraf

Promises vs callbacks: Promises are slightly less performant but nicer to work with. (We could add callbacks, too, if the need be.)

[joliss]

Sync has problems.

I think when you're seeing the async code work better, it's not because of any fundamental reason why async is more robust, but rather because async enables rimraf to do icky setTimeout-based workarounds that hide the problem rather than fixing it. I don't think that enabling these workarounds is sufficient reason to make quick-temp async. (Also see https://github.com/joliss/broccoli/issues/39#issuecomment-37486171.)

Isaac also recommends using async

The rimraf README is rather cryptic, but I assume that this recommendation is because the sync version doesn't have said workarounds. If you like you can ping Isaac about this and find out more. For now, I'm not willing to merge this PR, but if there turns out to be a deeper problem with the sync code, I'd be happy to revisit this.

[MajorBreakfast]

Hmm. Okay, I release it under https://github.com/MajorBreakfast/slick-temp :)

While I agree that it's bad that node's sync file functions throw more errors I suspect the reason for that is that they're not the way to go and as such haven't gotten enough developer attention. Blocking the main thread just isn't right. While at the moment with the rise of SSDs we've gotten pretty fast file systems I suspect that gap will widen again. For CPUs price and speed play a role. For memory price, speed and capacity play a role. I say that async code is more future proof. Using all that sync code also goes a bit against node's nature in general.