Add model.node_table()

[visr]

With the add API the full Node table is never created in memory, but the rows are available per node type. Users that need just the network need a way to compute the table when needed.

Marking this as a draft because I'm stuck. The new function works fine, but I tried re-enabling the round trip test, and found that the read from #1243 creates empty node tables for each node type due to the way it filters. I wanted to fix that by setting node.df = None if the filtered Node table was empty (if the node type doesn't occur).

But that triggers behavior I do not understand.

from ribasim import Model, Node
from ribasim.nodes import basin
from shapely.geometry import Point

model = Model(
    starttime="2020-01-01",
    endtime="2020-01-20",
)

model.basin.add(Node(5, Point(0, 0)), [basin.State(level=[1.0])])
# now both model.basin.node and model.basin.state have 1 row
model.basin.add(Node(8, Point(3, 0)), [basin.State(level=[2.0])])
# now model.basin.state has 2 rows, but model.basin.node only the last node 8

The setattr in the add method triggers this, so if you pass no tables like basin.State, model.basin.node contains both rows. I rewrote the if/else in add a bit but no real changes, the logic there seems ok. We just seem to trigger something pydantic that I find hard to debug, so any help welcome.

This is all triggered by my change to filter, so perhaps if there are better ways to fix that, by e.g. pruning empty node types at the end of model reading or something like that, I'm also happy.

[visr]

Closing in favor of a working PR #1279 and an issue #1278.