`_get_unique_label_values_as_index()` gives the wrong dtype

[LucaMarconato]

• in the example below, _get_unique_label_values_as_index() and this causes a mismatch during a join operation
• in general, I think we should raise a warning during the table validation if we detected a mismatch in the type between the type of the labels (or the type of the index of shapes/points) and the type of the instance_key column.

##
##
from anndata import AnnData
from spatialdata.datasets import blobs
from spatialdata.models import TableModel

from spatialdata._core.query.relational_query import _get_unique_label_values_as_index
from spatialdata import join_sdata_spatialelement_table

sdata = blobs()
name = "blobs_labels"
instance_id = _get_unique_label_values_as_index(sdata[name]).to_numpy()
n = len(instance_id)
new_table = AnnData(shape=(n, 0), obs={"region": [name for _ in range(n)], "instance_id": instance_id})
new_table = TableModel.parse(new_table, region=name, region_key="region", instance_key="instance_id")
sdata['new_table'] = new_table

##
print(sdata[name].dtype) # gives int16
print(new_table.obs['instance_id'].dtype) # gives int64

joined_elements, new_table = join_sdata_spatialelement_table(
 sdata=sdata, spatial_element_name=name, table_name='new_table', how="inner", match_rows="left"
)
print(new_table) # gives None, should not be None

CC @melonora

[LucaMarconato]

This is due to the parser, from the lines

if adata.obs[instance_key].dtype != int:
 try:
 warnings.warn(
 f"Converting `{cls.INSTANCE_KEY}: {instance_key}` to integer dtype.", UserWarning, stacklevel=2
 )
 adata.obs[instance_key] = adata.obs[instance_key].astype(int)
 except IntCastingNaNError as exc:
 raise ValueError("Values within table.obs[] must be able to be coerced to int dtype.") from exc

[LucaMarconato]

The error was caused by using the inner join for labels, and I was not seeing the raised warning because I was in a test. Anyway I have added, with #468, a function to validate the dtypes. The function is not called by TableModel.validate(), but we could call it for instance when reading/writing a SpatialData object.