Enable support for multiple graph-based decomposition with qp.decompose

[maliasadi]

Context:
Follow up to #2552

A ptototype for fully integrate the graph-decomposition with qp.decompose enabling support for multiple decompose transformations and replace the old (decompose-lowering) logic from the frontend. This will be revisited when the graph-decomposition pass offers feature parity with qp.decompose

Example:

@qml.register_resources({qml.RX: 1})
def x_to_rx(wire: int):
    qml.RX(np.pi, wire)

@qml.register_resources({qml.RY: 1})
def y_to_ry(wire: int):
    qml.RY(np.pi, wire)

@qml.register_resources({qml.RX: 1, qml.RY: 1})
def h_to_rx_ry(wire: int):
    qml.RX(np.pi / 2, wire)
    qml.RY(np.pi / 2, wire)


@qml.qjit
@qml.decompose(gate_set={qml.Rot})
@qml.transforms.merge_rotations
@qml.decompose(
    gate_set={qml.RX, qml.RY},
    fixed_decomps={qml.PauliX: x_to_rx, qml.PauliY: y_to_ry},
    alt_decomps={qml.H: [h_to_rx_ry]},
)
@qml.transforms.cancel_inverses
@qml.qnode(qml.device("lightning.qubit", wires=2))
def circuit(x: float, y: float):
    qml.H(0)
    qml.H(0)
    qml.RX(x, wires=0)
    qml.PauliX(0)
    qml.RY(y, wires=0)
    qml.PauliY(0)
    qml.RY(x + y, wires=0)

    return qml.state()

Benefits:

• Users can use qp.decompose instead of catalyst.graph_decomposition for multiple-decomposition applications

Possible Drawbacks:

Related GitHub Issues:
[sc-116956]