`template_domain_names` missing in features.pkl on multimer mode.

I tried to check out `features.pkl` generated by running AlphaFold to find which templates are used.

When `--model_preset=monomer` is used I could find `template_domain_names` as a key name of the dictionary unpickled, but not upon `--model_preset=multimer`.

```python
import pickle

with open("../features.pkl", "rb") as f:
features = pickle.load(f)
print('template_domain_name' in features) # => False

```

```python
features.keys()

# => `template_domain_names` not found
# dict_keys(['aatype', 'residue_index', 'seq_length', 'msa', 'num_alignments', 'template_aatype', 'template_all_atom_mask', 'template_all_atom_positions', 'asym_id', 'sym_id', 'entity_id', 'deletion_matrix', 'deletion_mean', 'all_atom_mask', 'all_atom_positions', 'assembly_num_chains', 'entity_mask', 'num_templates', 'cluster_bias_mask', 'bert_mask', 'seq_mask', 'msa_mask'])

```

When running with multimer mode, templates are not used for prediction models? (I could see things like template names in docker log file, though.)

Otherwise, running on multimer mode generates no data related to templates used for prediction while it makes use of templates' information?