Fix series length ordering for string[python] IDs in dataframe validation/conversion

[dario-fumarola]

Summary

Fixes #440 by making series-length extraction deterministic and aligned with row order, including when id_column uses pandas string[python] dtype.

Root cause

After sorting by (id_column, timestamp_column), the code used:
value_counts(sort=False).to_list()
to derive per-series lengths. For some ID dtypes (notably string[python]), this can produce an order that does not match contiguous row blocks, which then misaligns timestamp slicing and can trigger false frequency inference failures.

Changes

• In validate_df_inputs, replaced:
  • df[id_column].value_counts(sort=False).to_list()
  • with df.groupby(id_column, sort=False).size().to_list()
• Applied the same fix in convert_df_input_to_list_of_dicts_input when validate_inputs=False for consistency.
• Added regression tests:
  • test_validate_df_inputs_accepts_string_python_ids_with_unequal_lengths
  • test_validate_df_inputs_has_consistent_metadata_for_object_and_string_python_ids
  • test_convert_df_with_validate_inputs_false_handles_string_python_ids

Validation

• pytest test/test_df_utils.py (36 passed)
• mypy src test (no issues)

Compatibility

No public API changes. Behavior is unchanged except for correcting dtype-dependent ordering/misalignment.