(#37) - update benchmarks to more accurately include compilation time

[nolanlawson]

Fixes #37.

Lots of changes in this PR, mostly revolving around improving the accuracy of the benchmarks and preparing us for a 1.0.3 release.

The good news is that optimize-js still provides a measurable benefit in Chrome, Firefox, and Edge. The bad news is that I can confirm Safari is a slight regression (~-1%), and with the new, more accurate benchmarks, we can only boast more modest improvements in various browsers (~20% in Chrome rather than ~57%).

Since my goal with optimize-js is more to provide research than to provide the tool itself, though, I'm happy to just report the data as I find it. Also there's nothing to stop us from continuing to test, continuing to tweak, and continuing to improve. :smiley:

Major changes in this PR

• More accurate perf measurements using marky and as described in https://github.com/nolanlawson/optimize-js/issues/37#issuecomment-270583578
• Drop Ember from the benchmark – it was causing the browser to crash due to a memory leak, and its reliance on jQuery made it hard to work with
• New bin/print-bundle-sizes.sh script to more easily generate these for the README
• Automatically delete added window keys between each test, delete scripts after use to reduce memory overhead
• Update Lodash and PouchDB libraries to latest of each