Ensure consistent return type of CLs-like / p-value-like values

[matthewfeickert]

We've introducing inconsistent return types for CLs-like / p-value-like values now in PR #1162. Previously, all CLs-like / p-value-like values were 0-d tensors

pyhf/src/pyhf/infer/__init__.py

Line 149 in 0e71f2f

# Ensure that all CL values are 0-d tensors

However, now we have things like hypotest returning 0-d tensors and things like pvalues and expected_pvalues returning floats only for NumPy and for all other backends returning 0-d tensors, yet in all the examples we're showing CLs-like / p-value-like values being returned.

Example: Run the docstring example from expected_pvalues for various backends and note the return type of CLs_exp_band[0].

import pyhf

for backend in ["numpy", "pytorch", "jax"]:
    print(f"\nbackend: {backend}")
    pyhf.set_backend(backend)
    model = pyhf.simplemodels.hepdata_like(
        signal_data=[12.0, 11.0], bkg_data=[50.0, 52.0], bkg_uncerts=[3.0, 7.0]
    )
    observations = [51, 48]
    data = observations + model.config.auxdata
    mu_test = 1.0
    asymptotic_calculator = pyhf.infer.calculators.AsymptoticCalculator(
        data, model, test_stat="qtilde"
    )
    _ = asymptotic_calculator.teststatistic(mu_test)
    sig_plus_bkg_dist, bkg_dist = asymptotic_calculator.distributions(mu_test)
    CLsb_exp_band, CLb_exp_band, CLs_exp_band = asymptotic_calculator.expected_pvalues(
        sig_plus_bkg_dist, bkg_dist
    )
    print(f"CLs expected band: {CLs_exp_band}")
    print(
        f"of type: {type(CLs_exp_band[0])} and shape {pyhf.tensorlib.shape(CLs_exp_band[0])}"
    )

gives

backend: numpy
CLs expected band: [0.0026062609501074576, 0.01382005356161206, 0.06445320535890459, 0.23525643861460702, 0.573036205919389]
of type: <class 'numpy.float64'> and shape ()

backend: pytorch
CLs expected band: [tensor(0.0026), tensor(0.0138), tensor(0.0645), tensor(0.2353), tensor(0.5730)]
of type: <class 'torch.Tensor'> and shape ()

backend: jax
CLs expected band: [DeviceArray(0.00260626, dtype=float64), DeviceArray(0.01382005, dtype=float64), DeviceArray(0.0644532, dtype=float64), DeviceArray(0.23525643, dtype=float64), DeviceArray(0.57303619, dtype=float64)]
of type: <class 'jax.interpreters.xla._DeviceArray'> and shape ()

We should try to come to a consensus on what the return type for a CLs-like / p-value-like value should be. From looking back at PR #944 and Issue #714, I think that the CLs-like values should be 0-d tensor as it allows for us to sidestep this difference in backend behavior (I believe this is the motivation for the current behavior). While conceptually it makes sense to have a p-value-like just be a float to emphasize the scalar nature, having it be a 0-d tensor make it appear to a user as still having scalar like behavior.