Major refactor to clean up internal code, enable full parallelism.

[dmitshur]

This is a major refactor that should preserve current behavior, but update internal code significantly.

The four main consequences are:

• No more bottlenecks due to legacy architecture that prevented all operations that can be executed in parallel from being executed in parallel. Without such artificial limitations, the time to calculate Go-Package-Store output is significantly reduced.

  In a sample Go workspace with 3000~ Go packages (with many, many outdated packages), the previous version of Go-Package-Store took:

  9 minutes 35.116430321 seconds

  The new version took:

  37.706725795 seconds

• When providing -godeps or -govendor flags, previously, it would only display updates for packages that you happened to have in your GOPATH already. If a package mentioned in Godeps.json or vendor.json wasn't available in your GOPATH workspace, updates could not be presented. This restriction is lifted.
• No more importing legacy packages that have "gist" in the name. This makes the internal code easier to understand, modify, and maintain. Especially for other people.
• The repo.Updater interface and its implementations are simplified. Its responsibility is now only to update the provided repository, not to ensure updated packages are hidden, etc. This should help with #51.

The general idea behind the refactor is similar to the related project https://github.com/shurcooL/gostatus/pull/27.

Depends on https://github.com/google/go-github/pull/255 being merged to avoid data races, but they're harmless (the value under a data race is not being used), so not blocking.

[dmitshur]

I'm going to merge this now.

I've been testing the new version locally, and everything seems to be functional and stable, I have not caught any regressions.

There are still some TODOs scattered in the code, they are mostly about cleaning up and better organizing the code; all the functionally important and correctness-critical TODOs have been resolved by now.

I will continue to work on addressing the remaining style and code organization issues, but that shouldn't block this otherwise incredible performance and code simplification improvement PR from being merged.

I know some people have commented that Go Package Store was slow, and it truly was. After this change, it's incredibly fast, as it should be. This is all a result of simplifying the code and doing the parallel pipeline processing right. And no more gists; I was finally able to get rid of those pesky things. :) Time to :ship: it!