`qp.allocate(1)`: using the qubit without index works but not formally supported

When using qp.allocate(1) to allocate a single qubit, calling qp.H(w) (without indexing w[0]) silently succeeds. This behavior is not guaranteed and relies on implementation details. PL / Catalyst should either formally support this pattern or emit a warning/error directing users to always index their qubit registers.

Example:

import pennylane as qp

@qp.qjit(capture=True)
@qp.qnode(qp.device("null.qubit", wires=2))
def circuit():
    with qp.allocate(1) as w:
        # This works but is not guaranteed:
        qp.H(w)

        # This is the correct, explicit form:
        # qp.H(w[0])

    return qp.state()

print(circuit())

Versions:

Name: pennylane
Version: 0.45.0.dev52
JAX version:             0.7.1
Catalyst version:        0.15.0.dev47

Start by reproducing the provided qp.allocate(1) and qp.H(w) example with the listed PennyLane and Catalyst versions. Trace how qp.allocate and the unindexed register reach capture and operator handling; done means the behavior is formally supported or a clear warning/error directs users to qp.H(w[0]), with coverage for the chosen outcome.