TimeSeries.to_dataframe(): Bad columns in output DataFrame when data is empty.

[felixchenier]

Describe the bug
When a TimeSeries has multidimensional data shape but first dimension is 0 (no data), its to_dataframe() method doesn't produce the right DataFrame columns.

To Reproduce
df = pd.DataFrame(columns=['Data0', 'Data1[0,0]', 'Data1[0,1]',
'Data1[1,0]', 'Data1[1,1]'])
df['Data0'] = np.array([])
df['Data1[0,0]'] = np.array([])
df['Data1[0,1]'] = np.array([])
df['Data1[1,0]'] = np.array([])
df['Data1[1,1]'] = np.array([])

ts = ktk.TimeSeries.from_dataframe(df)
assert ts.data['Data0'].shape == (0,)
assert ts.data['Data1'].shape == (0, 2, 2)

df2 = ts.to_dataframe()
assert np.all(df == df2)

Expected behavior
df and df2 should be identical.