Bug in utils/torch.py base-case handling and lack of unit test coverage

**Description**
While implementing a new test suite for the utility modules, I identified a bug in `perceptionmetrics/utils/torch.py` specifically within the `get_data_shape()` function.

Currently, this function recursively traverse nested lists and tuples but assume that any non-sequence base case is a PyTorch tensor. This causes an `AttributeError` when the data structure contains metadata such as strings (filenames), integers (labels), or dictionaries.

**Bug Details**
In `get_data_shape()`:
```python
# Current implementation
elif torch.is_tensor(data):
return tuple(data.shape)
else:
return tuple(data.shape) # Raises AttributeError if data is a string/int
```
The `else` block should implement a "passthrough" logic (matching `data_to_device()`) to return the non-tensor data as-is, allowing for mixed-type data structures.

**Proposed Changes**
1. **Fix `torch.py`:** Update the `else` block in `get_data_shape()` to `return data`.
2. **Add Unit Tests:** The utility modules (`torch.py` and `image.py`) currently lack rigorous testing. We should implement a formal unit testing suite using `pytest` to validate recursive operations and visualization helpers to prevent future regressions.

**Impact**
Fixing this will enable the evaluation pipeline to handle datasets that bundle metadata (like image paths or class names) alongside tensors without crashing during shape validation.