[WIP] Simplify and fix one-hot encoding

[kristofgazso]

The library currently seems to be working with floats for one-hot encoding, and was also using for for loops for it, which is not very parallelizable

This doesn't seem usual, since categorical data points should normally be represented as ints as opposed to floats, as they can be expected to be used as indices

One test is still failing: test_point_counter_iris, maybe because of some weird float use. The final assert in the test seems shows that something is wrong:

old
poisoned                 120
not_poisoned             120
correctly_defended        54
incorrectly_defended       3
original_points_total    120
training_points_total    183

new
poisoned                 281
not_poisoned             -41
correctly_defended       240
incorrectly_defended       0
original_points_total    120
training_points_total    181

would be cool if the devs took a look =))