Tracking record's `get_dir` doesn't extend to cached runs

[WarmCyan]

Haven't fully tested, but it looks like a directory created with record.get_dir doesn't get included in a later store full run (using cached data) because the stage doesn't execute/get_dir doesn't run that second time.

[WarmCyan]

A potential partial solution to this might actually be to save a metadata file for the stage itself, which could include any get_dir/get_path calls. This would at least solve storing a previous run, it wouldn't impact DAG setup or the --paths flag from #131