Partialling errors of derived inputs in tt_function

[MImmesberger]

What problem do you want to solve?

This issue tracks the partialling error of inputs that would be derived functions in the tt_function. Here's an example:

• The policy_input is x_m
• You pass x_y as a scalar when requesting the tt_function
• Then, the tt_function will still ask for x_y as an input, i.e. the partialling that should have happened in step 2 did not happen

The reason is connected to the problems we had when building templates: Derived functions are not created when there is no input data to create them from. In particular, this line in with_processed_params_and_scalars is problematic, because there is no node x_y when creating the tt_function in the example above:

    for n, f in without_tree_logic_and_with_derived_functions.items():
        if n in processed_data:
             ... # This is where the partialling is prepared

Similarly, partialling doesn't work with inputs of already vectorized functions.

Originally raised in PR #54.