[Feature, Testing] Allow nested calls to ConcretizedCallable, more tests

[jpmoutinho]

Extends the ConcretizedCallable to allow nested calls, adds some convenience functions. E.g. now we can write

import pyqtorch as pyq

## Creating an expression and embedding

expr = "z" ** pyq.log(1.0 / (1.0 + (2.0 * pyq.sin("x")) + "y"))

expr_name = "expr"

embedding = pyq.Embedding(
    fparam_names = ["x", "y"],
    vparam_names = ["z"],
    var_to_call = {"expr": expr}
)

## Passing the expression to a block parameter

op = pyq.Scale(pyq.I(0), "expr")

values = {"x": torch.tensor(1.0), "y": torch.tensor(-1.0),  "z": torch.tensor(2.0)}

matrix = op.tensor(values, embedding)

matrix[..., 0]

---

tensor([[0.6971, 0.0000],
        [0.0000, 0.6971]])

Furthermore, with the current change, we might not need the Embedding 🤔. To be explored in a later MR: #278

[jpmoutinho]

Thanks @vytautas-a :)

I think my main question is regarding the logic on how to write HamiltonianEvolution, and which are the correct combinations of parameters to write a time-dependent one, time-independent one, etc.

[jpmoutinho]

Thanks @vytautas-a, almost there but I think we should keep the usage of ConcretizedCallable as a parameter in the Parametric as well for consistency, since you added it to the Scale.