Improve URI module efficiency

[fastjames]

The Stripe.URI module used a less-efficient method for building the HTTP query params string. Here, we switch to a more efficient list prepend and reverse pattern. Also, the recursive parser made use of Enumerable.impl_for/1 which is fairly heavyweight. By checking for maps or lists (which are assumed to be keyword lists) we can remove impl_for. I wish I could DRY up that check, but the small difference between the guard and non-guard syntax prevents it.

Addresses #105

[coveralls]

Coverage decreased (-1.04%) to 24.074% when pulling 23f79f7 on geminimvp:uri_efficiency into f6ddf08 on code-corps:2.0.

[coveralls]

Coverage decreased (-1.7%) to 23.423% when pulling 7613a42 on geminimvp:uri_efficiency into f6ddf08 on code-corps:2.0.

[joshsmith]

@DavidAntaramian @SViccari would you mind reviewing?

[joshsmith]

@asummers this definitely goes away with your changes being pulled in, right?

[asummers]

It would. Any performance problems would be the purview of the dependency library.

[joshsmith]

@fastjames have you looked at this URI library we're talking about at all? Any indication it suffers some of the same problems?

[joshsmith]

@fastjames I'm closing since it's outdated (we're using a dependency library), but thanks for the contribution. Let me know if you have an answer to my previous question at all.

[fastjames]

@joshsmith sorry for the delay -- it looks like the dependency library uses reduce |> reverse, so it should be fine.
https://github.com/shhavel/uri_query/blob/master/lib/uri_query.ex#L35

[joshsmith]

Thanks for the update @fastjames!