the problems with the original implementation was mostly around error
handling. sources had to be tracked manually and provided to each
incremental output. the `hasArtifact` check was frequently forgotten.
this has been re-abstracted through `incr.work()`, which is given an
`io` object. all fs reads and module loads go through this interface,
which allows the sources to be properly tracked, even if it throws.
closes#12
chloe caruso
66da129036