Expand Hera's capacity to deserialize results on behalf of users

[flaviuvadan]

Is your feature request related to a problem? Please describe.
It's challenging for users to know when they need to JSON deserialize an output of a step/task in Argo.

Describe the solution you'd like
Hera should have the ability to infer what elements to deserialize based on the specification of a workflow.

Describe alternatives you've considered
Not doing this at all.

Additional context
Here's an example that illustrates the problem. At a high level:

• we have a workflow that runs a set of steps
• one of the steps is a reusable set of steps that is called on the output of a script that generates some numbers
• we want to collate the outputs of those independent steps
• the independent output of the reusable steps pipeline gets JSON encoded by Argo, resulting in independent integers coming in as JSON values
• Hera successfully deserializes the JSON list that contains the values but does not deserialize the independent values

@script(image="python:3.10")
def generate_starting_values():
    print(list(range(1, 4)))


@script(image="python:3.10")
def echo(value: int):
    print(value)


@script(image="python:3.10")
def double(value: int):
    doubled = value * 2
    # sleep(30 * (value + 1))
    print(doubled)


@script(image="python:3.10")
def square(value: int):
    squared = value * value
    # sleep(30 * (value + 1))
    print(squared)


@script(image="python:3.10")
def all_sum(values: list):
    print(sum(map(int, values)))  # <<<<<<<<<<<<< we have to use `map` here to turn from str to int


@script(image="python:3.10")
def all_echo(values: list):
    print(values)


with Workflow(
    generate_name="callable-steps-",
    entrypoint="calling-steps",
) as w:
    with Steps(
        name="steps",
        inputs=Parameter(name="value"),
        outputs=Parameter(name="values", value_from=m.ValueFrom(parameter="{{steps.square.outputs.result}}"))
    ) as my_steps:
        e = echo(arguments={"value": "{{inputs.parameters.value}}"})
        d = double(arguments={"value": e.result})
        square(arguments={"value": d.result})

    with Steps(name="calling-steps") as s:
        g = generate_starting_values()
        m = my_steps(arguments={"value": "{{item}}"}, with_param=g.result)
        
        with s.parallel():
            all_sum(arguments={"values": m.get_parameter("values").value})
            all_echo(arguments={"values": m.get_parameter("values").value})

w.create()