fix add_edge_to_view leaving edge_id to -1 in SQL

[TeunHuijben]

Fix: add_edge_to_view leaves the local edge_id as -1 on a SQL root

Reviving a soft-deleted edge into a view with add_edge_to_view left the view's local copy of the edge with edge_id = -1 (the placeholder) when the root is a SQLGraph. The id map was updated correctly, so view.edge_id(a, b) returned the right root id while view.edge_attrs() still showed -1 for that row — the two disagreed, and any read of that edge's attributes by edge id hit zero rows.

Reproduction

import polars as pl
from tracksdata.graph import SQLGraph

g = SQLGraph(drivername="sqlite", database=":memory:")
g.add_edge_attr_key("weight", pl.Float64)
a = g.add_node({"t": 0})
b = g.add_node({"t": 1})
edge_id = g.add_edge(a, b, {"weight": 1.5})

view = g.filter().subgraph()
view.remove_edge_from_view(a, b)   # hide the edge in the view only
view.add_edge_to_view(a, b)        # bring it back

print(view.edge_id(a, b))                      # 1    -> id map is correct
print(view.edge_attrs()["edge_id"].to_list())  # [-1]  -> local row is wrong
print(view.edges[edge_id]["weight"])           # ValueError

ValueError: can only call '.item()' if the Series is of length 1, or an explicit
index is provided (Series is of length 0)

The root graph is fine throughout, and the in-memory backends are unaffected: there the view reuses the root's edge payload by reference, so the id comes along for free. _add_edge_local only breaks for a non-rx root, where it re-fetches the edge attrs, drops edge_id, and never stamps it back.

Fix

Stamp the root edge id into the reconstructed payload in _add_edge_local, the same way add_edge and bulk_add_edges already do.

Test

test_add_edge_to_view_keeps_edge_id — revive an edge into a view, then assert the id map, the edge_attrs() row, and a read by edge id all agree. Parametrized over all three backends; it failed on SQLGraph only before the fix.