Allow to evaluate cost consequences of different scaling decisions

[jonnor]

For a particular config, or comparing two configs, what are the expected effects in terms of costs.
Should be seen in relationship with #45 (analyzing perf on real data).

Also interesting for evaluating code changes, like #33.

[jonnor]

Some (experimental) code in ./src/analysis.coffee

[jonnor]

Maybe the default usage could be to to show estimated cost, compared with the currently active config? At the end should just be the grand total, with +- difference (and %). There should also be a similar breakdown per-role, so one can evaluate where differences come from.

Significant differences should possibly be highlighted with a warning color yellow/orange.

When storing scaling config in git, one can use git show to get alternative versions: git show HEAD~4:autoscale.guv.yaml

[jonnor]

One might also want to have validation rules, that says one cannot go above a certain level in costs.