Simplify tests generated by erfa_generator.py

[eerovaher]

Among other things erfa_generator.py translates the erfa tests in t_erfa_c.c into a Python module that can be tested with pytest. So far the translation has tried to change the code as little as possible, but writing Python as if it is ANSI C is not taking advantage of the features Python and pytest provide. A less literal translation removes the need for the viv() and vvd() test helpers and improves tracebacks if a test should fail. For example, with the following change to the erfa/tests/test_ufunc.py generated by current main

@@ -3729,7 +3729,7 @@ def test_xys06a():
     x, y, s = erfa_ufunc.xys06a(2400000.5, 53736.0)
     vvd(x,  0.5791308482835292617e-3, 1e-14, "xys06a", "x", status)
     vvd(y,  0.4020580099454020310e-4, 1e-15, "xys06a", "y", status)
-    vvd(s, -0.1220032294164579896e-7, 1e-18, "xys06a", "s", status)
+    vvd(s, -0.1220032294164579896e-7, 0.0, "xys06a", "s", status)
 
 
 @pytest.mark.xfail(np.__version__ < '1.24', reason='numpy < 1.24 do not support no-input ufuncs') 

the resulting traceback is

_____________________________________________ test_xys06a _____________________________________________

    def test_xys06a():
        x, y, s = erfa_ufunc.xys06a(2400000.5, 53736.0)
        vvd(x,  0.5791308482835292617e-3, 1e-14, "xys06a", "x", status)
        vvd(y,  0.4020580099454020310e-4, 1e-15, "xys06a", "y", status)
>       vvd(s, -0.1220032294164579896e-7, 0.0, "xys06a", "s", status)

erfa/tests/test_ufunc.py:3732: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

val = np.float64(-1.22003229416848e-08), valok = -1.22003229416458e-08, dval = 0.0, func = 'xys06a'
test = 's', _ = array(0)

    def vvd(val, valok, dval, func, test, _):
        """Validate a double result."""
        a = val - valok
>       assert a == 0.0 or abs(a) <= abs(dval), (
            f"{func} failed: {test} want {valok:.20g} got {val:.20g} "
            f"(1/{abs(valok / a):.3g})")
E       AssertionError: xys06a failed: s want -1.2200322941645799674e-08 got -1.2200322941684799586e-08 (1/3.13e+11)
E       assert (np.float64(-3.899991152065015e-20) == 0.0 or np.float64(3.899991152065015e-20) <= 0.0)
E        +  where np.float64(3.899991152065015e-20) = abs(np.float64(-3.899991152065015e-20))
E        +  and   0.0 = abs(0.0)

erfa/tests/test_ufunc.py:41: AssertionError

but if an analogous change is made to the file generated by this PR

@@ -3713,7 +3713,7 @@ def test_xys06a():
     x, y, s = erfa_ufunc.xys06a(2400000.5, 53736.0)
     assert x == pytest.approx(0.5791308482835292617e-3, abs=1e-14)
     assert y == pytest.approx(0.4020580099454020310e-4, abs=1e-15)
-    assert s == pytest.approx(-0.1220032294164579896e-7, abs=1e-18)
+    assert s == pytest.approx(-0.1220032294164579896e-7, abs=0.0)
 
 
 @pytest.mark.xfail(np.__version__ < '1.24', reason='numpy < 1.24 do not support no-input ufuncs')

then the traceback is much shorter and much clearer:

_____________________________________________ test_xys06a _____________________________________________

    def test_xys06a():
        x, y, s = erfa_ufunc.xys06a(2400000.5, 53736.0)
        assert x == pytest.approx(0.5791308482835292617e-3, abs=1e-14)
        assert y == pytest.approx(0.4020580099454020310e-4, abs=1e-15)
>       assert s == pytest.approx(-0.1220032294164579896e-7, abs=0.0)
E       assert np.float64(-1...229416848e-08) == -1.2200322941...e-08 ± 0.0e+00
E         
E         comparison failed
E         Obtained: -1.22003229416848e-08
E         Expected: -1.22003229416458e-08 ± 0.0e+00

erfa/tests/test_ufunc.py:3716: AssertionError