bfloat16 floating-point data type.

NumPy does not have built-in support for bfloat16. As an extension, TensorStore defines the tensorstore.bfloat16.dtype NumPy data type (also available as numpy.dtype("bfloat16"), as well as the corresponding tensorstore.bfloat16.type array scalar type, and these types are guaranteed to interoperate with TensorFlow and JAX.