compile training-relevant exaspim datasets

To better coordinate work on training models for reconstruction correction predictions, we want to create combined data assets and/or DocDB queries that point to all available samples with complete reconstructions from the v2 scope (and possibly also the v1 scope separately). This will likely involve some updates to the metadata, and will need support from Peter and maybe Anna.

not strictly for the NMCP but that's the closest milestone.

Start by reviewing the available DocDB queries and metadata for complete reconstructions in the v2 scope, then check whether v1 should be handled separately. Coordinate with Peter and possibly Anna on the required metadata and training assets; done means all in-scope samples are discoverable through combined assets and/or queries.

Written by the indexing model from the issue text.