Process CSV import one row at a time to permit back-references

[glennmatthews]

Closes: #4998

What's Changed

Instead of processing all rows of a CSV in a single ListSerializer instance, process each row sequentially. This may (may) result in a performance hit, but it permits self-referential models such as TreeModels to be defined in a CSV where later rows reference back to objects defined/created in earlier rows, which was previously not possible in a single CSV import, e.g. for LocationTypes:

name,nestable,parent__name
Region,TRUE,NoObject
Site,FALSE,Region
Building,FALSE,Site

TODO

• Explanation of Change(s)
• Added change log fragment(s) (for more information see the documentation)
• Attached Screenshots, Payload Example
• Unit, Integration Tests
• n/a Documentation Updates (when adding/changing features)
• n/a Example Plugin Updates (when adding/changing features)
• Outline Remaining Work, Constraints from Design

[HanlinMiao]

Is there any performance impact for this change?

[glennmatthews]

Is there any performance impact for this change?

Almost certainly. I'll see if I can gather some data to quantify it.

[glennmatthews]

Importing 71 IPAddress records via CSV took approximately 15 seconds in develop and approximately 19 seconds in this branch, both with DEBUG = False to turn off DjDT. So it's slower, but not excessively so IMHO.

We definitely should plan to move CSV import to a background task / system job, in any case.