Add an option to cache results for subsequent runs

[timstclair]

For large projects, it would be nice to not need to rebuild the dependency graph on each run.

In its simplest form, this could be 2 additional flags for saving the graph to an output file and reading the graph from a saved file.

A more advanced implementation might be able to automate this detecting file modifications, and only rescanning the modified packages.

[evvvvr]

Not sure about this second part with detecting changes and scanning only modified packages because IMO the way to do this is:

1. Have info about each package imports in cache
2. Gather info about imports for each package
3. If info for package imports from step 2 doesn't match info from step 1 — package imports has been changed

But this involves package graph walk, which we tried to avoid.

[timstclair]

I think the second option should be possible by looking at modification times (on unix -- not sure if this is possible on windows), and only rescanning the imports from that package if a file changed (or was added / removed). I haven't thought through the details though.