Ufunc of NDVarArray

[IgnaceBleukx]

Helene exposed an issue with our implementation of the NDVarArray overridden functions (mul, add....) in pull request #337

To reproduce:

import cpmpy as cp

w = cp.cpm_array([10,20,30])
x = cp.intvar(0,10, shape=3)

w *x 
>> NDVarArray(NotImplemented,NotImplemented,NotImplemented)

What happens is the __mul__ method of NDVarArray w tries to call __mul__ on each of the elements in that array. I.e., 10.__mul__(x[0]), 20.__mul__(x[1])... but the built-in type int has no idea how to multiply with our custom objects.
Instead of calling __rmul__ on our objects as it does when expressing 10 * x[0], a NotImplementedError is returned.

Did some research on how to properly subclass a np.ndarray and it seems we don't really need the _vectorized thingy...
https://numpy.org/doc/stable/reference/arrays.classes.html

[363734]

This is the code that triggered the discovery: tentative of refactoring the decompose method of the cumulative cst:

for t in range(lb,ub+1):
            demand_at_t = 0
            for job in range(len(start)):
                if is_num(demand):
                    demand_at_t += demand * ((start[job] <= t) & (t < end[job]))
                else:
                    demand_at_t += demand[job] * ((start[job] <= t) & (t < end[job]))
            cons += [capacity >= demand_at_t]

to

for t in range(lb, ub+1):
            cons += [capacity >= sum(demand * ((start <= t) & (t < end)))]

[Wout4]

this issue also mentions ufunc of ndvararray, so might also be solved? #124

[Wout4]

another issue that might be fixed here: #71
can we support something like x[x!=2] (like numpy array does)