[python] Control sets cannot mix a qubit with a qview, and cudaq.control fails on kernels that call other kernels

[wsttiger]

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.

Describe the feature

Two related gaps make "one external control qubit + an ancilla register"
control sets awkward in Python kernels:

1. Gate level: any mix of a bare qubit and a qview in one control set is
   rejected (invalid argument type for control operand). All-qubit and
   single-view control sets work.
2. Kernel level: cudaq.control(kernel, control, args...) works for leaf
   kernels (including list arguments) but fails with "Could not
   successfully apply kernel specialization" when the controlled kernel
   itself calls another kernel — so composite operations cannot be
   controlled wholesale.

import cudaq
cudaq.set_target("qpp-cpu")

# (1) mixed control set
@cudaq.kernel
def mixed():
    c = cudaq.qubit()
    q = cudaq.qvector(3)
    x.ctrl(c, q.front(2), q[2])   # error: invalid argument type for control operand

# (2) cudaq.control of a composite kernel
@cudaq.kernel
def leaf(q: cudaq.qview):
    for i in range(q.size()):
        x(q[i])

@cudaq.kernel
def composite(q: cudaq.qview):
    leaf(q)

@cudaq.kernel
def caller():
    c = cudaq.qubit()
    q = cudaq.qvector(2)
    x(c)
    cudaq.control(leaf, c, q)       # OK
    cudaq.control(composite, c, q)  # RuntimeError: Could not successfully apply kernel specialization.

Workaround: co-allocate the control qubit and the ancillas in one register
and slice views (register.front(...)/register.back(...)) — but that
leaks into public APIs: users must co-allocate their control qubit with a
library's ancilla register instead of passing any qubit they own (e.g. a
QPE counting qubit). Either fixing (1) or (2) would remove the constraint;
(2) alone would allow cudaq.control(subroutine, ctrl, ...) over composite
operations.

Environment

• CUDA-Q built from source at commit 0be565550f4c23affdcbed9e4eaec38d2d0915e6
• Python 3.11.x, target qpp-cpu (library mode)
• AlmaLinux 8.10, x86_64

All reproducers below were executed and verified against this build.