custom gate register_operation doesn't seem to work if it is invoked in one kernel which is nested in another.

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

cudaq.register_operation appears not to work in a nested kernel situation. see code example below

Steps to reproduce the bug

my_matrix = np.matrix([[0, 1], [1, 0]])
cudaq.register_operation("my_gate", my_matrix)

@cudaq.kernel
def my_kernel(qubits : cudaq.qview):
    my_gate(qubits[0])
    
@cudaq.kernel
def kernel(num_qubits : int):
    qubits = cudaq.qvector(num_qubits)
    my_kernel(qubits[0]) # Can't call custom gate through another kernel here
    my_gate(qubits[0]) # Can call the custom gate directly here 

result = cudaq.sample(kernel, 2, shots_count=1000)
print(result)

Expected behavior

Expected to call the my_gate through the my_kernel call.

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• **CUDA-Q version **: 0.9.1
• Python version:
• C++ compiler:
• Operating system:

Suggestions

No response

Start with the cudaq.register_operation API and reproduce the provided nested-kernel example on CUDA-Q 0.9.1. Verify the behavior through cudaq.sample: the custom my_gate should work both directly in kernel and through my_kernel without changing the reported expected behavior.