Expand SWEEP_LIKE to work for {'t': [0, 1, 3]}

Currently if you make a call like sampler.sample(circuit, params={'t': range(5)}) it doesn’t work, because it expects e.g. a list of dictionaries instead of a dictionary containing lists. For convenience we should just support both cases. Check if the input is a dictionary with list values (and possibly singletons too), and if so take the cartesian product of the various entries and turn it into a sweep.

Test cases:

assert cirq.to_sweeps({'t': [0, 2, 3]}) == (
    cirq.to_sweeps([{'t': 0}, {'t': 2}, {'t': 3}]}))

assert cirq.to_sweeps({'t': [0, 1], 's': [2, 3], 'r': 4}) == (
    cirq.to_sweeps([
        {'t': 0, 's': 2, 'r': 4},
        {'t': 0, 's': 3, 'r': 4},
        {'t': 1, 's': 2, 'r': 4},
        {'t': 1, 's': 3, 'r': 4},
    ]))

[kevinsung]

Note that Sampler.sample actually does use to_sweeps, so it does work with this feature. So as it stands, this feature is inconsistently implemented.

[maffoo]

I think we decided not to do this so that you can resolve list-valued parameters (we have use cases for this internally). Also, there’s an ambiguity if you have multiple parameters; given {'a': [1, 2], 'b': [3, 4]} should that do a cartesian product sweep over a and b or a zip sweep? Instead you need to make the sweep explicit at the top level, either as [{'t': 0.3}, {'t': 0.5}] or using actual sweep objects: cirq.Points('t', [0.3, 0.5]).