Prevent invalid utf8 indexing in cell magic detection (#9146)
The example below used to panic because we tried to split at 2 bytes in
the 4-bytes character `转`.
```python
def sample_func(xx):
"""
转置 (transpose)
"""
return xx.T
```
Fixes #9145
Fixes https://github.com/astral-sh/ruff-vscode/issues/362
The second commit is a small test refactoring.
