Improve custom unitary operations

Required prerequisites

• Search the issue tracker to check if your feature has already been mentioned or rejected in other issues.

Describe the feature

Improvement suggestions for custom unitary operations:

• Allow unitary operations to be passed a qvector as a parameter
• Support kernel mode.
• Support lookup of custom unitary ops with parameterized strings (see #2137 for details)

Example of code where those features would be beneficial:

import cudaq
import numpy as np

@cudaq.kernel
def kernel_helper(state: list[complex]):
    qubits = cudaq.qvector(state)

def flip_qubit(matrix, i):
    j = i - 1
    idx = 1 << i
    if (j < 0):
        jdx = 0
    else:
        jdx = 1 << j
    matrix[jdx][jdx] = 0
    matrix[idx][idx] = 0
    matrix[idx][jdx] = 1
    matrix[jdx][idx] = 1
    return matrix

def test_unitary():
    def register_custom_operation(unitary_index: int, matrix: np.ndarray):
        cudaq.register_operation(f"U_{unitary_index}", matrix)

    state = np.zeros(8)
    state[0] = 1
    state = np.array(state, dtype=cudaq.complex())
    states = [state]
    for i in range(0,3):
        matrix = flip_qubit(np.identity(8), i)
        register_custom_operation(i, matrix)

        kernel, s = cudaq.make_kernel(cudaq.State)
        qubits = kernel.qalloc(s)
        kernel.__getattr__(f'U_{i}')(qubits[0], qubits[1], qubits[2])

        cudaq_state = cudaq.State.from_data(state)
        state = np.array(cudaq.get_state(kernel, cudaq_state))
        states.append(state)

    for s in states:
        counts = cudaq.sample(kernel_helper, s)
        print(counts)

test_unitary()

# Output:

# { 000:1000 }
# { 001:1000 }
# { 010:1000 }
# { 100:1000 }

Start by reproducing the example using cudaq.register_operation, cudaq.make_kernel, and cudaq.sample, then trace how custom unitary operations are registered and invoked. Review the behavior required for qvector parameters, kernel mode, and parameterized operation lookup as described in the issue and #2137. Done means the listed use cases work and the example produces the expected states.