Fix test failing due to Numpy 2.0 promotion rules

From Numpy 2.0 adding a numpy.float32 and a Python numeric type returns
a numy.float32 when it previously returned a numpy.float64. This changes
the behavior when using the Python builtin sum function on a
numpy.float32 array as the internal computations now will be performed
as numpy.float32 additions when it used to be numpy.float64.

Passing a numpy.double(0) as a start value to the innermost sum forces
the old behavior.

python/test/segyio_c.py

@@ -540,10 +540,10 @@ def read_line(f, metrics, iline_idx, xline_idx):
     buf = numpy.zeros((len(iline_idx), samples), dtype=numpy.single)
 
     f.getline(xline_trace0, len(iline_idx), xline_stride, offsets, buf)
-    assert sum(sum(buf)) == approx(800.061169624, abs=1e-6)
+    assert sum(sum(buf), numpy.double(0)) == approx(800.061169624, abs=1e-6)
 
     f.getline(iline_trace0, len(xline_idx), iline_stride, offsets, buf)
-    assert sum(sum(buf)) == approx(305.061146736, abs=1e-6)
+    assert sum(sum(buf), numpy.double(0)) == approx(305.061146736, abs=1e-6)
 
     f.close()