Support for Rank-0 (input None)

I am trying to convert microsoft/trocr-small-handwritten (PyTorch)

Link to Colab Notebook: https://colab.research.google.com/drive/1fuvyNc63c2NXy0TjB6T3fGSC6wLZPJwC?usp=sharing (Ignore the ONNX Section)

I can successfully trace both the encoder and decoder models, and even successfully export the encoder model, but the decoder model errors out with: `ValueError: Rank-0 (input None) is unsupported`

```
import coremltools as ct

ct_enc_model = ct.convert(traced_encoder, inputs=[ct.ImageType(shape=xenc.shape)])
```

```
ct_dec_model = ct.convert(traced_decoder,
inputs=[
ct.TensorType(shape=input_ids.shape),
ct.TensorType(shape=attention_mask.shape),
ct.TensorType(shape=encoder_hidden_states.shape),
ct.TensorType(shape=pad_size.shape)
])

WARNING:coremltools:Tuple detected at graph output. This will be flattened in the converted model.
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
[](https://localhost:8080/#) in
----> 1 ct_dec_model = ct.convert(traced_decoder,
2 inputs=[
3 ct.TensorType(shape=input_ids.shape),
4 ct.TensorType(shape=attention_mask.shape),
5 ct.TensorType(shape=encoder_hidden_states.shape),

10 frames
[/usr/local/lib/python3.8/dist-packages/coremltools/converters/mil/mil/program.py](https://localhost:8080/#) in __init__(self, sym_shape, dtype, name, allow_rank0_input)
161 if len(sym_shape) == 0:
162 if not allow_rank0_input:
--> 163 raise ValueError('Rank-0 (input {}) is unsupported'.format(name))
164 else:
165 logger.warning('Rank-0 (input {}) is unsupported in coreml. You might run into error while\

ValueError: Rank-0 (input None) is unsupported
```

The reason I have labelled this as a question is because I am unsure if not supporting Rank-0 is a bug or if this is just unsupported by coremltools altogether.

Reproduce the decoder conversion with the linked Colab notebook, ignoring its ONNX section, then inspect coremltools/converters/mil/mil/program.py around the allow_rank0_input check shown in the traceback. Determine whether rank-0 inputs can be supported or are a Core ML limitation; done means the issue has a documented resolution and, if support is feasible, verified conversion behavior.

Written by the indexing model from the issue text.