Make edge_id user configurable

[evetion]

Fixes #1718

Sets node_id and edge_id as primary key fields in the geopackage. Also sets fid indexes for all non-node/edge tables. This now sorts the Node table by node_id only (used to be node_type first).

I triggered some bugs by the amount of code I touched, so I changed some more:

• Split contextvars into reading and writing (arrow file was trying to be read on writing a model).
• Simplified sorting
• Moved all the DataFrame(dict(**kwargs)) boilerplate into a single validator on TableModel.

[visr]

Am I correct that this is only breaking in the sense that the new version expects edge_id in Edge, whereas old models don't have that. But reading and writing with Python will fix that? Can we read old models with Python or will it fail on the presence of a fid column?

[evetion]

Am I correct that this is only breaking in the sense that the new version expects edge_id in Edge, whereas old models don't have that. But reading and writing with Python will fix that? Can we read old models with Python or will it fail on the presence of a fid column?

Correct on both. You can read older models with Python, as pyogrio hardcodes the index to fid anyway, we rename it afterwards. 😄

[visr]

Regarding this point from #1718:

The non-spatial tables (Basin / static etc) also get an fid, that falls outside our data model but was needed to make the tables editable in QGIS. This was done in #873. To avoid confusion, Marnix tried to hide these columns in #841 (comment) but ran into a QGIS bug that has since been fixed. So we should try again to hide these columns for non-spatial tables.

Should we create an issue to start hiding fid columns in QGIS again? Am I correct that fid is now carried over, an no longer reset each time, by Ribasim Python? But it still isn't really part of the data model.

A bit of a nit, but in QGIS node_id shows up as the first column, but edge_id as the last. Can we make edge_id the first as well?

This id column is hardcoded in the iMOD QGIS plugin to fid, but that doesn't affect anything other than a confusing name:

https://github.com/Deltares/imod-qgis/blob/afa0eb694d367999c328e0dffc45f0bafb8230d1/imodqgis/timeseries/timeseries_widget.py#L471

[evetion]

Should we create an issue to start hiding fid columns in QGIS again? Am I correct that fid is now carried over, an no longer reset each time, by Ribasim Python? But it still isn't really part of the data model.

I tried to make things consistent by having the indexes always there in Python with the same name as well, not just in geopackage. While the fid index resets during normal use (sorting when writing destroys the original fid), it should carry over to and from QGIS. We can make an issue, if you dislike seeing the fid, but I don't care that much either way.

A bit of a nit, but in QGIS node_id shows up as the first column, but edge_id as the last. Can we make edge_id the first as well?

Not sure why that happens, in the table schema it shows up as the first column.

[visr]

Great, let's merge then. The fid in the tables don't bother me that much either. edge_id on the left would be nice, but perhaps not under our control then.