Support dask 2025.4.0

Dask 2025.4.0 optimizes multiple DataFrames together, which exposes
division mismatches when assigning a pandas Series to a dask DataFrame
column. The old reset_index/set_index workaround no longer avoids this.
Replacing it with compute-assign-rewrap via dd.from_pandas, which builds
a clean expression graph. This is safe because __getitem__ already
computes the DataFrame to produce the Series being assigned.

Author: filippsatverily

cdisc_rules_engine/models/dataset/dask_dataset.py

@@ -81,9 +81,9 @@ def __setitem__(self, key, value):
             array_values = da.from_array(value, chunks=tuple(chunks))
             self._data[key] = array_values
         elif isinstance(value, pd.Series):
-            self._data = self._data.reset_index()
-            self._data = self._data.set_index("index")
-            self._data[key] = value
+            pdf = self._data.compute()
+            pdf[key] = value.values
+            self._data = dd.from_pandas(pdf, npartitions=1)
         elif isinstance(value, dd.DataFrame):
             for column in value:
                 self._data[column] = value[column]

pyproject.toml

@@ -15,7 +15,7 @@ dependencies = [
     "cachetools >=6.1.0",
     "cdisc-library-client >=0.1.6",
     "click >=8.1.7",
-    "dask[dataframe,array] >=2024.6.0, <2025.4.0",
+    "dask[dataframe,array] >=2024.6.0",
     "fastparquet >=2024.2.0",
     "importlib-metadata >=8.5.0",
     "jsonata-python >=0.6.0",