[FIX] contingency.pyx: Fix out of bound write

[ales-erjavec]

Issue

Computing contingencies for more then 127 classes results in an out or bounds write in contingency_floatarray possibly leading to a segmentation fault (in the best case) or just wrong results.

import Orange.statistics.contingency, numpy
X = Orange.data.ContinuousVariable("X")
C = Orange.data.DiscreteVariable("C", values=["C{}".format(i + 1) for i in range(1024)])
domain = Orange.data.Domain([X], [C])
d = Orange.data.Table.from_numpy(
    domain,
    numpy.random.uniform(size=(4000, 1)).round(3),
    numpy.random.randint(0, 1024, size=(4000, 1)),
)
c = Orange.statistics.contingency.get_contingency(d, X, C)

consistently results in a segfault on macOS Python 3.5 due to memory corruption.

Description of changes

• Enable cython bounds check on contingency_floatarray.
• Change dtype of classes to intp_t to allow more then 127 'class' values.

Includes

• Code changes
• Tests
• Documentation

[codecov-io]

Codecov Report

Merging #2924 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2924      +/-   ##
==========================================
+ Coverage   82.14%   82.14%   +<.01%     
==========================================
  Files         328      328              
  Lines       56417    56417              
==========================================
+ Hits        46343    46344       +1     
+ Misses      10074    10073       -1

[lanzagar]

Nice catch!