Provide a custom repr for function params in `@step` decorator

I have a function like:
```
@step
def foo(x: np.ndarray, y: Dict[str, np.ndarray]): ...
```
It is marked with `@step` decorator and my problem is that the output in the allure repo is too large. I see that the special case of `represnet` is here https://github.com/allure-framework/allure-python/blob/3c74fd540021f776be7b52f8078c9d34eee5c3b6/allure-python-commons/src/utils.py#L167

So I was wondering how to handle other objects like `np.ndarray` or `Dict[str, np.ndarray]`?

One way is to subclass `dict` and give a nice `repr` to it, but what with a single numpy array?Sublassing it just so it got a better `repr` for allure looks... excessive?

Can `@step` accept an additional argument that will explicitly tell it hot to handle `obj -> str` representation?

#### I'm submitting a ...
- [x ] feature request

#### What is the current behavior?

A long text output in the allure report for types like `np.ndarray` or `Dict[str, np.ndarray]` that have long `repr`.

#### What is the expected behavior?

Make `@step` that receives explicit formatter so I can write:

```
def _my_repr(x: np.ndarray, y: Dict[str, np.ndarray]):
return format_1(x), format_2(t) # <-- custom formatting code here

@step(formatter=_my_repr)
def foo(x: np.ndarray, y: Dict[str, np.ndarray]): ...
```

#### Please tell us about your environment:

allure-pytest==2.9.45
allure-python-commons==2.9.45
pytest==6.2.5

#### Other information

I am willing on implementing that feature conditional on the fact there is a way of accepting that change to the mainstream. Will you accept a PR that parametrize `@step` in such a way?

Start with the special-case representation logic in allure-python-commons/src/utils.py and trace how @step collects function parameters. Define how an explicit formatter should be accepted and applied to parameters, then verify that numpy arrays and dictionaries produce the requested compact output without changing default behavior.

Written by the indexing model from the issue text.