feat: initialize inputs in IImageInference post-init

[rensortino]

Summary

When IImageInference is constructed with inputs passed as a plain dict (e.g. {"referenceImages": [...]}) instead of an IInputs instance, the SDK later calls _addOptionalField which invokes inputs.to_request_dict() — a method that only exists on the IInputs dataclass. This results in an AttributeError at runtime.

Adds a __post_init__ hook to IImageInference that automatically converts a raw dict value for inputs into a proper IInputs instance via IInputs(**self.inputs), matching the pattern already used by IInputs itself for its own __post_init__ logic.

Related issue

Fixes #244

Problem

Passing inputs as a dictionary when constructing IImageInference — a natural pattern when building request params dynamically — fails because the dataclass stores the raw dict rather than an IInputs object:

request = IImageInference(
    positivePrompt="...",
    model="bfl:7@1",
    inputs={"referenceImages": [image_data_uri]},  # plain dict → crash
)

This raises an AttributeError: 'dict' object has no attribute 'to_request_dict' during serialization in _addOptionalField.

Solution

A __post_init__ method on IImageInference detects when inputs is not None and not already an IInputs instance, and coerces it:

def __post_init__(self):
    if self.inputs is not None and not isinstance(self.inputs, IInputs):
        self.inputs = IInputs(**self.inputs)

This allows users to pass either an IInputs object or a plain dictionary, with both working seamlessly.

[Sirsho1997]

Thanks @rensortino
Will be adding such post_init() for all other taskTypes as well.

[Sirsho1997]

Check this #246