Fix cardinality and integer cast

[bouthilx]

Fix cardinality

There was two errors.

1. The integer bounds are inclusive, but the cardinality was computed as if
   the upper bound was exclusive.

2. The shape affects cardinality exponentially, it would be
   (cardinality of one item) ** prod(shape), not (cardinality of one item) * prod(shape).

Fix casting of Integer

Why:

There is a mismatch between the inclusive bound of Integer and
the underlying distribution (which is real) where high is in theory exclusive
but in practice inclusive due to limited precision and rounding.
We cannot use numpy.round() to fix this because otherwise the probability
is changed for lowest and highest possible numbers. For instance a
uniform distribution (0, 10) would have a probability of 0.5 * (1/10)
for 0 instead of (1/10).

How:

In the cast method, rescale the point from (low, high) to
(low, high + 1) before computing the numpy.floor().astype(int).
Note that some distributions have infinite intervals, in which case we
do no rescaling.

Deprecate randint distribution

uniform(a, b, discrete=True) and randint(a, b) handles differently the intervals are
seems incompatible. We were already recommending uniform over randint
anyway.