Fix ValueError in FillMissingParameters with empty experiment data (#4973)

Summary:
Pull Request resolved: #4973

  `FillMissingParameters.transform_experiment_data` crashes with
  `ValueError: Columns must be same length as key` when `arm_data` has
  zero rows (e.g., during initial Sobol trial generation before any
  trials have completed).

  This happens because `DataFrame.apply(func, axis=1)` on an empty
  DataFrame returns an empty DataFrame rather than an empty Series.
  Assigning that DataFrame to a single column then fails.

  Replace `arm_data.apply(...)` with a list comprehension over
  `arm_data.iterrows()`, which correctly produces an empty list for
  empty DataFrames.

Failing run: https://www.internalfb.com/mlhub/flow/1044918711/overview

Reviewed By: yuhuishi-convect, saitcakmak

Differential Revision: D95266027

Author: Richa Gadgil

ax/adapter/transforms/fill_missing_parameters.py

@@ -101,10 +101,7 @@ def transform_experiment_data(
                 if c != "metadata" and c not in self._derived_parameters
             ]
             for p_name, p in self._derived_parameters.items():
-                arm_data[p_name] = arm_data.apply(
-                    lambda row, p=p: p.compute({col: row[col] for col in tunable_cols}),
-                    axis=1,
-                )
+                arm_data[p_name] = p.compute_array(arm_data)
         return ExperimentData(
             arm_data=arm_data,
             observation_data=experiment_data.observation_data,