[Opinfo] Better ErrorMsg for test_out with wrong shape

[ejguan]

While I implement Opinfo for inner #55536, the SkipInfo is added due to AssertionError: UserWarning not triggered : Resized a non-empty tensor but did not warn about it.

We assume the output tensor with wrong shape will be resized into expected shape

pytorch/test/test_ops.py

Lines 614 to 627 in b1d17bc

	def _case_two_transform(t):
	wrong_shape = list(t.shape)
	if len(wrong_shape) == 0:
	# Handles scalar tensor case (empty list)
	wrong_shape = [2]
	else:
	wrong_shape[-1] = wrong_shape[-1] + 1
	return make_tensor(wrong_shape, dtype=t.dtype, device=t.device)
	out = _apply_out_transform(_case_two_transform, expected)
	msg_fail = "Resized a non-empty tensor but did not warn about it."
	with self.assertWarnsRegex(UserWarning, "An output with one or more elements", msg=msg_fail):
	op_out(out=out)

But torch.inner doesn't resize the output tensor as we expected, even if it's a bug (reported #56022).

>>> a = torch.randn(5)
>>> b = torch.randn(5)
>>> torch.inner(a, b)
tensor(-4.2595)
>>> c = torch.randn(2)
>>> torch.inner(a, b, out=c)
tensor([-4.2595, -4.2595])

We should have a better error message in this case like raising an error of Output tensor is not resized as expect rather than assertion for resizing.

cc @mruberry @VitalyFedyunin @walterddr

[mruberry]

Thank you for filing this issue, @ejguan. So torch.inner() broadcasts to out=!? This is a great bug. And thank you for filing an issue for the test and the operation.