Support sampling

[josh146]

With the latest development version of PennyLane, we now support returning samples from the QNode, as well as expectation values and variance:

@qml.qnode(dev)
def circuit(x):
    qml.RX(x, wires=0)
    qml.CNOT(wires=[0, 1])
    return expval(qml.PauliZ(0)), sample(qml.PauliZ(1))

Plugins can support this new feature by subclassing the Device.sample() method, returning the requested number of samples for a given observable.

Note: some of the observables (qml.X, qml.P, qml.Homodyne) correspond directly to Measure operators in Strawberry Fields. This should be the proper way to support sampling in PennyLane-SF.

However, there are some SF Measurements that currently don't correspond to a PL observable, namely MeasureFock() for PNR measurements. Perhaps we could utilize qml.FockStateProjector(), and make n an optional argument?

Another issue is that there are some PL observables that don't correspond to a SF Measurement - qml.NumberOperator, qml.PolyXP. Since these are Gaussian observables, we should be able to support them in simulator backends (since we have access to the Gaussian state).