See title. We will have a _unminimized corpus directory which can contain millions of files and we sync a small batch continuously over time, merge, etc.

One fix landed - https://chromereviews.googleplex.com/523987014/.

Still need to fix some GCS handing with unicode. E.g. 300 million+ corpus from skia has a whole bunch of unicode char filenames. also, chars like [, ], etc are special chars in gcs, need to be escaped.