Slow for large databases

[icemac]

@jimfulton wrote at 2018-03-25:

I think this is going to be prohibitively slow for large databases. The required downtime will be too long.
For automated fixups, I think it would be better to work at the pickle level. I'm thinking of a tool that converts individual pickles. (This might be useful for analysis as well.)
This would have a lot of advantages:

• Potentially much faster because it doesn't require instantiating objects or running application-level code.
• Doesn't require application classes to be around.
• Can be integrated with existing workflows:
  • export/import
  • zodbcopy
  • ZRS replication.

I'll make a basic version of this that others can build on.