Revise returning empty dataframes from data_for_parameter

```
def data_for_parameter(ensemble: Ensemble, key: str) -> pd.DataFrame:
group, _ = _extract_parameter_group_and_key(key)
try:
df = ensemble.load_scalars(group)
except KeyError:
return pd.DataFrame()

dataframe = df.to_pandas().set_index("realization")
dataframe.columns.name = None
dataframe.index.name = "Realization"
data = dataframe.sort_index(axis=1)
if data.empty or key not in data:
return pd.DataFrame()
data = data[key].to_frame().dropna()
data.columns = pd.Index([0])
return data.astype(float)
```

Should instead propagate error up and handle it at callsite, and return an appropriate http error through dark storage which should be handled on the frontend (PlotAPI)

Start at data_for_parameter and trace its callers to identify where empty DataFrames are currently handled. Then follow the dark storage HTTP path and the PlotAPI frontend handling. Done means the underlying error propagates to the callsite, dark storage returns an appropriate HTTP error, and PlotAPI handles that response.

Written by the indexing model from the issue text.