Update docs/tutorials/main_tutorial.ipynb

Nobody has claimed this yet.

Under the new registration paradigm, directly accessing configurations via `hssm.defaults.default_model_config["model_name"]` might not work if `"model_name"` is not _registered_ beforehand.

Relatedly, there are a few instances of data from this dict being pulled to build other objects:
```python
angle_loglik = make_likelihood_callable(
loglik="angle.onnx",
loglik_kind="approx_differentiable",
backend="jax",
params_is_reg=[0, 0, 0, 0, 0],
)

ANGLE = make_distribution(
"angle",
loglik=angle_loglik,
list_params=hssm.defaults.default_model_config["angle"]["list_params"], # to update
)
```
This would work is the model is already registered, but it might be better to use the new functions that render these objects:
```python
from hssm.defaults import get_angle_config
angle_list_params = get_angle_config()["list_params"] # new interface for defaults

angle_loglik = make_likelihood_callable(
loglik="angle.onnx",
loglik_kind="approx_differentiable",
backend="jax",
params_is_reg=[0, 0, 0, 0, 0],
)

ANGLE = make_distribution(
"angle",
loglik=angle_loglik,
list_params=angle_list_params,
)
```

After implementing #659, `default_model_config` _could_ be converted into a defaultdict leveraging this future functinality. However, this feels a bit too overengineered. It might just be better and simpler to use these new functions explicitly and hide `default_model_config`.

@AlexanderFengler @digicosmos86

Start in docs/tutorials/main_tutorial.ipynb and search for direct accesses to hssm.defaults.default_model_config. Review the examples involving angle and replace those accesses with the corresponding new defaults functions, such as get_angle_config; done means the tutorial no longer relies on the old interface and its examples remain usable.

Written by the indexing model from the issue text.