Add bitwise elementwise specifications

[kgryte]

This PR

• adds bitwise elementwise functional specification equivalents for the following operators: &, |, ^, ~, >>, and <<.
• is derived from comparing API signatures across array libraries.

Notes

• The specifications follow the same form as other elementwise functions.

• Functional bitwise APIs are, while not ubiquitous, commonly implemented across array libraries. The principle impetus for their inclusion is to have functional equivalents of operators (as discussed and endorsed during a prior consortium meeting).

• This PR induces a slightly different naming convention. NumPy (and its followers) splits its namespace into non-prefixed and prefixed APIs (e.g., right_shift vs bitwise_and). TensorFlow puts everything under a tf.bitwise namespace, while also using a bitwise_ prefix for certain APIs.

  This PR chooses to simply put all bitwise operations under a bitwise_ prefix. This matches, and is consistent with, the current convention of using a logical_ prefix for logical operations.

• This PR currently requires that x2 for both right_shift and left_shift have nonnegative elements. NumPy's docs for left_shift state that this is required; although, from the interpreter, no error is thrown if a negative x2 is provided. TensorFlow states that negative x2 results in implementation-dependent behavior.

  This PR takes the stance that negative x2 for either left_shift and right_shift is probably a mistake and should not be permitted. Hence, the included requirements that x2 be greater than or equal to 0.

[rgommers]

LGTM, thanks @kgryte. I like the choices made and the consistency of the resulting API here. Since this deviates a little from existing implementations, I'll leave it open for at least a few days in case there are more comments.

[kgryte]

@rgommers I've updated the bitwise_left_shift and bitwise_right_shift function names per the discussion in the previous consortium meeting. This PR should be ready for merge.

[rgommers]

Thanks @kgryte, merged.