`config` structure

Nobody has claimed this yet.

> Since have nested and rather complex config structure, this approach is not suitable. You should have a custom class for this using `dataclass` for example, since working with nested dict objects is difficult and not maintainable. I know you're in a time crunch and want to produce results as soon as possible. But I highly recommend refactoring this part. So, for example, instead of having `config['real']['results']`, you can have something like this:

```python
@dataclass
class RealModel:
inp: Path
graph: Path
subcatchments: Path
results: Path

def __post_init__(self)->None:
self.inp = Path(self.inp)
self.graph = Path(self.inp)
self.subcatchments = Path(self.inp)
self.results = Path(self.inp)
for f in (self.inp, self.graph, self.subcatchments, self.results):
if not f.exists():
raise FileNotFoundError(f)
@dataclass
class ModelConfig:
real: RealModel
```

> So, whenever you want to use it in the code, you can use `config.real.results`. You can have validations and conversions, if necessary, in `__post_init__`. Having a structure for config even makes development and maintenance easier, since most IDEs now use them for autocomplete suggestions and static type checkers will be able to catch errors.

> As I said, if you don't have time to refactor this now, it's ok, it's not a blocker for me.

_Originally posted by @cheginit in https://github.com/ImperialCollegeLondon/SWMManywhere/pull/83#discussion_r1526522626_

No files, tests, or entry points are named. Locate where the nested configuration is constructed and accessed, then assess the proposed Python dataclass structure and its validation needs. Done means configuration access no longer depends on the difficult-to-maintain nested dictionary structure and existing behavior remains covered by the project's tests.

Written by the indexing model from the issue text.