Use pytest.raises instead of the raises argument to pytest.mark.xfail

This is more appropriate for the use case.

From the pytest docs:
Using pytest.raises is likely to be better for cases where you are
testing exceptions your own code is deliberately raising, whereas using
@pytest.mark.xfail with a check function is probably better for
something like documenting unfixed bugs (where the test describes what
"should" happen) or bugs in dependencies.

c.f.
https://docs.pytest.org/en/latest/assert.html#assertions-about-expected-exceptions

tests/test_tensor.py

@@ -4,10 +4,10 @@
 from pyhf.simplemodels import hepdata_like
 
 
-@pytest.mark.xfail(raises=KeyError)
 def test_astensor_dtype(backend):
     tb = pyhf.tensorlib
-    assert tb.astensor([1, 2, 3], dtype='long')
+    with pytest.raises(KeyError):
+        assert tb.astensor([1, 2, 3], dtype='long')
 
 
 def test_simple_tensor_ops(backend):