Support mutable datasets

[jonathanthiry]

Mutable datasets would be especially useful for near real-time data. They could use UUIDs instead of PIDs based on the file's checksum.
Both files could be streamed at the same time to check that data has only been appended and that the new file includes at least the same data as the old one.

[DrMaggie]

Keeping the same PID for the changed digital object would not really be a good idea if the identifier string is constructed from the checksum of the original version. The setup we have now with assigning new PIDs to later versions (regardless of the reason for the version change - i.e. treating files with new appended values the same as those where there's been error corrections or similar - is much clearer. There is always the possibility to create identifiers that are always pointing to the "latest version" of data sets that we know a priori will be extended or otherwise updated on a regular basis- but then the associated PID strings should not be based on any fixity metadata! Related to this issue, we could also think of ways to display the reason(s) behind a version being created, as well as make it easy to see the full version history, on landing pages?

[jonathanthiry]

Thanks, Maggie! Oleg also thought it would be better to use another identifier, like UUIDs, so we will do that.