Don't enforce a non-empty shape on tensors

[matthewfeickert]

Description

As @lukasheinrich pointed out in PR #944, enforcing a shape of (1,) on tensors created from floats is probably not a good idea when none of the backends do this

>>> import pyhf
>>> import numpy as np
>>> 
>>> example = np.asarray(0.1)
>>> print(f"example {example} is a {type(example)} with shape {example.shape}")
example 0.1 is a <class 'numpy.ndarray'> with shape ()
>>> pyhf.set_backend("numpy")
>>> example = pyhf.tensorlib.astensor(0.1)
>>> print(f"example {example} is a {type(example)} with shape {example.shape}")
example [0.1] is a <class 'numpy.ndarray'> with shape (1,)

So we should make sure that we match the shape here

# make this so
assert np.asarray(0.1).shape == pyhf.tensorlib.astensor(0.1).shape

[matthewfeickert]

It seems that we implemented this behavior in PR #413 because of MXNet. 🙃 Given that we don't support MXNet anymore I think there's no reason that we shouldn't implement @lukasheinrich's suggestion, as I think Issue #288 won't pop up again.

[matthewfeickert]

@lukasheinrich As it affects PR #972, you would agree that pyhf.infer.test_statistics.qmu should return a () scalar and not a (1,) tensor, yes?