Fix bin1d_vec to properly account for floating point precision

The tolerance for bin size `h` must be based on eps and the _involved_ numbers in calculating it, not the resulting number. The two involved numbers are of the same order of magnitude, and one if it is `a0` (the first bin edge `bin[0] == min(bins)`), for which we already calculate the tolerance (`a0_tol`). So we can use `a0_tol` also as tolerance for `h`.

csep/utils/calc.py

@@ -80,7 +80,7 @@ def bin1d_vec(p, bins, tol=None, right_continuous=False):
         h = bins[1] - bins[0]
 
     a0_tol = numpy.abs(a0) * numpy.finfo(numpy.float64).eps
-    h_tol = numpy.abs(h) * numpy.finfo(numpy.float64).eps
+    h_tol = a0_tol  # must be based on *involved* numbers
     p_tol = numpy.abs(p) * numpy.finfo(numpy.float64).eps
 
     # absolute tolerance