Add searchsorted to the specification

[kgryte]

This PR

• resolvesRFC: add searchsorted to the specification #688 by adding a specification for searchsorted to the Array API specification.
• requires that x1 be a one-dimensional array. PyTorch and TensorFlow allow x1 to be multi-dimensional; however, NumPy et al do not. In this PR, I've chosen to restrict x1 to a one-dimensional array, as this is the lowest common denominator. The specification can be extended in the future to accommodate multi-dimensional x1 (stacking) in the future.
• requires that x2 be an array. CuPy and TensorFlow require that x2 be an array, while NumPy, PyTorch, and others allow x2 to be a scalar. In this PR, I've chosen to restrict x2 to be an array. For the scalar use case, downstream consumers can wrap scalar values as zero-dimensional arrays.

[rgommers]

Thanks @kgryte. I have a few minor comments, but +1 overall to this addition.

The issue is linked to from a scikit-learn PR where searchsorted was needed, and SciPy needs it as well (Cc @mdhaber for visibility).

[kgryte]

@rgommers This PR should be ready for final sign off.

[rgommers]

Reviewed again, all comments were addressed and this PR looks good to me. I'll plan to merge it later today or tomorrow.

[asmeurer]

In this PR, I've chosen to restrict x1 to a one-dimensional array, as this is the lowest common denominator. The specification can be extended in the future to accommodate multi-dimensional x1 (stacking) in the future.

Shouldn't the language be "x1 should be a one-dimensional array" instead of "x1 must be a one-dimensional array"?

[asmeurer]

Should we only require this to be the default array index type?

[kgryte]

Why would that be necessary?