Wrong result from d2dtf() in pre-1972 UTC

[mrhso]

Related: #76 #82

#76 was fixed in SOFA 17a, however, it has brought another problem.

      leap = (fabs(dleap) > 0.5);

The threshold is 0.5 s, which means that mini-leaps, of which the maximum absolute value is 0.1077580 s, cannot be processed properly.

>>> from astropy.time import Time
>>> Time('1965-06-30 12:00:00', format = 'iso', scale = 'utc')
<Time object: scale='utc' format='iso' value=1965-06-30 11:59:59.950>

Considering that the minimum absolute value of UTC mini-leaps is 0.005 s (is there such a serious rounding error?), it is possible to determine an appropriate threshold.