[python] cudaq.sample API assumes broadcast on nested list arguments

Should cudaq.sample be changed to accept an explicit broadcast parameter?

Details:

cudaq.sample checks if the first argument is a 2-d array and broadcasts by passing elements of the array to the kernel.

This leads to confusion for a non-broadcasting sample that expects a nested list, for example:

@cudaq.kernel
def test_list_list(a: list[list[int]]):
    qc = cudaq.qvector(3)
    j = 0
    for i in a:
        if i[0] == 1:
            x(qc[j])
        j += 1
    mz(qc)

print(cudaq.sample(test_list_list, [[0,1],[1,0],[1,0]]))

As a workaround, we can make kernel take a non-list first parameter:

@cudaq.kernel
def test_list_list(n: int, a: list[list[int]]):
    qc = cudaq.qvector(3)
    j = 0
    for i in a:
        if i[0] == 1:
            x(qc[j])
        j += 1
    mz(qc)

print(cudaq.sample(test_list_list, 2, [[0,1],[1,0],[1,0]]))

Related: https://github.com/NVIDIA/cuda-quantum/issues/1809
Related: https://github.com/NVIDIA/cuda-quantum/issues/2627

Start at the cudaq.sample entry point and reproduce the provided test_list_list example, comparing it with the workaround that adds a leading integer parameter. Trace how nested-list arguments are classified for broadcasting. Done means the intended broadcast behavior is explicit and the nested-list case is covered by a regression test.