ENH: np.roll: extra argument to pad with zeros or constant

Proposed new feature or change:

In many situations, it is required to shift an array (like np.roll) and fill the values with zeros, instead of making the array "loop".

Is there a way to implement this:

x = np.array([1, 2, 3, 4, 5]
np.roll(x, 2, method="constant")   #  [ 0 0 1 2 3 ]   instead of  [ 4 5 1 2 3 ]
np.roll(x, -2, method="constant")  #  [ 3 4 5 0 0 ]  instead of   [ 3 4 5 1 2 ]

?

Example : python numpy roll with padding

Important notes:

• solutions with np.pad exist but are slower, see https://stackoverflow.com/questions/2777907/python-numpy-roll-with-padding#comment3270288_3153267

• other people have the needs for this: https://stackoverflow.com/questions/2777907/python-numpy-roll-with-padding#comment120688507_2778068: I think np.roll should do it natively with a keyword argument. (quite upvoted)

• this solution works: https://stackoverflow.com/a/2778068 but it would be interesting for the Numpy API to have this useful option (in the same way np.pad has constant mode, etc.)

Start with the np.roll API and review how its current wrapping behavior is implemented. Compare the requested zero and constant-fill examples, then determine the API and behavior changes needed; done means the new option works for positive and negative shifts without changing existing np.roll behavior.