Do not GC source data files

[dmpetrov]

dvc gc can remove not reproducible files that were added by dvc add. The only way to recreate the file:

dvc remove data/file.txt
cp ~/Download/file.txt data
dvc add data/file.txt 

We need to distinguish reproducible and not-reproducible (data file and source data file) files and not to remove the not-reproducible without a special option.

[efiop]

I would imagine this happening when you were switching between branches and calling gc, which removed cache that was used in a previous branch, but not being used in a current one.

Losing 'source data' is indeed an inconvenience... We should use our local state db to mark those cache files as 'source' ones and only remove them when gc is called with the flag --all or smth like that.

I will take a closer look shortly.

[dmpetrov]

Why local state? How about keeping this info in stage file?

[efiop]

Because stage files change when you switch branches, and local state db stays the same

[efiop]

Leaving for the next release.