remove np random generators for memory reasons

[maximelucas]

@maximelucas The combinatorial explosion of the mask array makes it impossible to use in some scenarios: e.g.

>>> import xgi
>>> xgi.random_hypergraph(1000, order=5, ps=[1e-12])  # => should create a network with ~1350 edges
numpy.core._exceptions._ArrayMemoryError: Unable to allocate 9.72 PiB for an array with shape (1368173298991500,) and data type float64

But that also has an easy solution of not pre-allocating mask:

import xgi
import numpy as np
import itertools

def random_hypergraph(N, ps, order=None, seed=None):
    rng = np.random.default_rng(seed)

    if order is not None:
        if len(ps) != 1:
            raise ValueError("ps must contain a single element if order is an int")

    if (np.any(np.array(ps) < 0)) or (np.any(np.array(ps) > 1)):
        raise ValueError("All elements of ps must be between 0 and 1 included.")

    nodes = range(N)
    hyperedges = []

    for i, p in enumerate(ps):

        if order is not None:
            d = order
        else:
            d = i + 1  # order, ps[0] is prob of edges (d=1)

        potential_edges = itertools.combinations(nodes, d + 1)
        edges_to_add = [e for e in potential_edges if rng.random() <= p]

        hyperedges += edges_to_add

    H = xgi.empty_hypergraph()
    H.add_nodes_from(nodes)
    H.add_edges_from(hyperedges)

    return H

This doesn't fail but this still takes $O({{N}\choose{m}})$ time to run.

Originally posted by @arashbm in #339 (comment)

Summary: go back to list comprehension rather than np.random()

[arashbm]

You can also chain a generator of np.randoms of a few millions or so to potentially get the best of both worlds.