Add support for broadcasting to `linalg.cross`

[kgryte]

This PR

• addresses https://github.com/data-apis/array-api/issues/415 by adding support for broadcasting to linalg.cross. Broadcasting follows NumPy behavior in only broadcasting the non-compute dimensions. With the exception of TensorFlow, this behavior is consistent (or will be after PyTorch addresses its broadcasting behavior) among array libraries.
• updates vecdot to match linalg.cross in being more explicit that broadcasting only applies to non-broadcast dimensions.
• brings linalg.cross in line with linear algebra design principles in which linear algebra APIs commonly support batching.