Dissociate CPN / OPN cleanly

Both the torch and the jax pipeline have the following decision rule applied for the `network_type` setting.

Torch:

```
if network_type is not None:
self.network_type = network_type
else:
self.network_type = "lan" if self.train_output_type == "logprob" else "cpn"
print(
'Setting network type to "lan" or "cpn" based on train_output_type. \n'
+ "Note: This is only a default setting, and can be overwritten by the network_type argument."
)
```

Jax:

```

# Identify network type:
if self.model.train_output_type == "logprob":
network_type = "lan"
elif self.model.train_output_type == "logits":
network_type = "cpn"
else:
network_type = "unknown"
print(
'Model type identified as "unknown" because '
"the training_output_type attribute"
' of the supplied jax model is neither "logprob", nor "logits"'
)

```

Anchoring on the `train_output_type` is the basic mistake. Since both the `cpn` and the `opn` networks have `train_output_type="logits"`.

This needs a slight refactor with new logic. Possibly not "identify"/"deduce" a name for the `network_type` and instead just pass that actively.

**Note**:

For backward compatibility, the previous (current) logic might need to be kept around.

Start by locating the torch and jax pipeline code that applies the `network_type` decision rules, then compare how `cpn` and `opn` models are represented when both use `train_output_type="logits"`. Define and verify the new explicit network-type flow, including the backward-compatibility requirement noted in the issue. Done means cpn and opn are no longer conflated by train_output_type in both pipelines.

Written by the indexing model from the issue text.