Lantern: Improve performance opportunity estimates based on graphs

[patrickhulce]

The potential savings estimates currently just compute the milliseconds based solely on bytes saved. This leads to some confusion and misrepresentation since all ms are not created equal, i.e. shaving 1s off FMP means a lot more than saving 1s off the end of "all images have finished downloading"

We can use the fast mode graphs to predict how much time would be saved for each metric as well as overall savings

[patrickhulce]

blocked on UX thinking here

[patrickhulce]

I am so stoked about how this is shaping up early. It's reranking things much more in line with how we intuitively think of biggest bang-for-buck performance. Improvements to files that are loaded early and dependencies of lots of resources now have the biggest impact whereas improvements to lazily-loaded images that are already as late as they should be have little to no impact.