Implement ObjectStep using a Function1

[kyri-petrou]

It seems we've gone a long way around to finally end up here, but I find it very difficult to believe that there is a more optimized version of this 😅

In short, this PR removes all the intermediate creation of Maps for us to create an ObjectStep, and instead uses a Function1. This allows us to:

1. Avoid wrapping ObjectSteps with a MetadataFunctionStep
2. Avoids creation of the intermediate Map[String, Step[R]]

In order to avoid breaking source-compatibility when creating ObjectStep, I also added a constructor that takes a Map[String, Step[R]]

The downside to this change is that in cases where we have aliased fields, the same field will be resolved more than once. However, in practise this is not really a problem because almost always aliased fields are then resolved to a FunctionStep. If we ever see that this might end up being an issue, we can add an intermediate cache, but I highly doubt it will be the case

[ghostdogpr]

I was thinking we could potentially remove Option and return NullStep directly to save the allocation. But you won't be able to use orElse, even though it can be worked around. Any thoughts?

[kyri-petrou]

Yeah I was thinking that as well but then hit the .orElse point. One way would be to check if the step is a NullValue but that could be a legit case of a field returning None.

Another would be using a PartialFunction instead, although I don't think it'd be as performant. I can give it a go

[kyri-petrou]

Yeah PartialFunction seems too messy. Returning NullStep and then merging fields as below works, but has the caveat I mentioned above

case (ObjectStep(name, fields1), ObjectStep(_, fields2)) =>
      ObjectStep(
        name,
        s =>
          fields2(s) match {
            case NullStep => fields1(s)
            case step     => step
          }
      )

[ghostdogpr]

Maybe it's okay since merging arbitrarily keeps one of them anyway (we're not supposed to have the same field twice in theory)?

[kyri-petrou]

Yeah I think you're right. The chances that this is a legit usecase are extremely low anyways. Changed fields to String => Step[R] instead

[ghostdogpr]

Squeezing the performance to the last bit 😄