Extend pydantic input model to hold primitive defaults

[liamhuber]

@samwaseda and I were chatting on how to handle actual input data in a recipe context. We don't have a conclusion on actual input values, but he did convince me to include primitive defaults.

I'm not sure how to modify the actual model structure to fit these in yet, but the rough idea is to permit a new default field per-input like default: OK where

from typing import Union

JsonPrimitive = str | int | float | bool | None
OK = JsonPrimitive | tuple[OK]  # Forgive the recursion, I'm just trying to log the point

Where deserialization converts the json list back to a python tuple.

Don't support mutable json-compliant stuff (list and dict) because actual uses of mutable defaults are almost exclusively going to be mistakes, and for the odd time it wouldn't be there are alternative approaches.

[liamhuber]

I've been thinking about how to implement this, and the direction I'm leaning right now is to replace inputs: Labels with inputs: list[InputPort] where

class InputPort(pydantic.BaseModel):
    label: Label
    default_value: OK | NotData = NotData # E.g., above
    has_python_default: bool = False

And NotData is a None-like value to indicate non-data (since None might actually be a legitimate default).

The has_python_default can be post-validated as always false except for AtomicNode and WorkflowNode, and there it gets validated against the presence of a fully_qualified_name (always present for atomic nodes, optionally present in workflow nodes). We can toggle it true in parse_atomic and parse_workflow when the source function is being examined.

What I like about this extra has_python_default field is that it lets us validate recipe input completeness (under the assumption that the recipe is not flat-out lying about having python defaults) and resolve #87 without have to actually hold and serialize arbitrarily complex default objects. I.e., a post-validation that for each node holding child nodes or prospective nodes, to verify the double loop for each (prospective) child for each input port that there is either a corresponding input edge or the port has a default value or is flagged as having a python default.