Call predict() for every user, every item

Nobody has claimed this yet.

Hello. I have been trying to use the LightFM predict function for every user and every item in the system. Essentially, my problems occur with supplying the input to predict, as I am unsure what kind of format user_ids and item_ids should take. My code is complex, so I'm going to supply simple examples.

Function definition:
`predict(user_ids, item_ids, item_features=None, user_features=None, num_threads=1)`

Suppose I want to get predictions for 10 users with ids [0, 1, 2, 3, 4, 5, 6, 7, 8, 9] and 10 items with ids [0, 1, 2, 3, 4, 5, 6, 7, 8, 9].

How would I use predict() to get a prediction for every user-item interaction? The following attempts do not work (and numerous others I have tried):

1)
```
user_ids = np.arrange(10)
item_ids = np.arrange(10)
predict(user_ids, item_ids)
```

2)
```
user_ids = np.arrange(10)
item_ids = [ np.arrange(10) for i in range(10) ]
predict(user_ids, item_ids)
```

I am able to use predict inside of a for loop, such as

```
user_ids = np.arrange(10)
item_ids = np.arrange(10)
for i in user_ids:
predict(i, item_ids)
```

However, I would prefer to avoid this, as I can only imagine that calculating everything at once would be more efficient.

Once I figure this out, I will work on incorporating the features.

Thank you for your help!

Start with the predict(user_ids, item_ids, item_features=None, user_features=None, num_threads=1) entry point and compare its expected input shapes with the two examples and the working loop. Done means clearly documenting how to request predictions for every user-item pair and how feature arguments fit into that usage.

Written by the indexing model from the issue text.