Relational query reorders table `obs` by region in 0.8

[timtreis]

Since #1131, filtering a table that annotates multiple regions reorders its obs rows by region instead of preserving the original order. This shifts table rows relative to element geometries and broke an ordering assertion in a spatialdata-plot test.

Cause: _filter_table_by_elements was changed from a boolean mask (order-preserving) to join_spatialelement_table(how="left"), which does obs.groupby(region) and thus regroups the rows.

MRE

import numpy as np
import pandas as pd
from anndata import AnnData
from geopandas import GeoDataFrame
from shapely.geometry import Point
import spatialdata as sd
from spatialdata.models import ShapesModel, TableModel


def circles(n):
    gdf = GeoDataFrame({"geometry": [Point(i, i) for i in range(n)], "radius": [1.0] * n})
    return ShapesModel.parse(gdf)


# table annotates two regions, obs rows INTERLEAVED: a,b,a,b,a,b
obs = pd.DataFrame(
    {
        "region": pd.Categorical(["a", "b", "a", "b", "a", "b"]),
        "instance_id": [0, 0, 1, 1, 2, 2],
        "label": ["a0", "b0", "a1", "b1", "a2", "b2"],
    }
)
table = TableModel.parse(
    AnnData(X=np.zeros((6, 1)), obs=obs, var=pd.DataFrame(index=["g0"])),
    region=["a", "b"],
    region_key="region",
    instance_key="instance_id",
)
sdata = sd.SpatialData(shapes={"a": circles(3), "b": circles(3)}, tables={"table": table})

before = list(sdata["table"].obs["label"])
after = list(sdata.filter_by_coordinate_system("global")["table"].obs["label"])
assert after == before, f"obs reordered {before} -> {after}"

Result

┌─────────┬───────────────────────────────────────┐
│ version │ obs after filter_by_coordinate_system │
├─────────┼───────────────────────────────────────┤
│ 0.7.2   │ a0, b0, a1, b1, a2, b2 (preserved) ✅ │
├─────────┼───────────────────────────────────────┤
│ 0.8     │ a0, a1, a2, b0, b1, b2 (regrouped) ❌ │
└─────────┴───────────────────────────────────────┘