Support BFloat16 dtype in Android Tensor API

[GregoryComer]

🚀 The feature, motivation and pitch

Currently, the ExecuTorch Android Java tensor class does not support BFloat16 dtype tensors. This prevents Java users from passing BFloat16 tensors in or out of the model. BFloat16 is increasingly common for GenAI, so this will be a useful feature to have.

To implement support, we will need to logic to Tensor.java. This task will require the following:

• Create a Tensor_bfloat16 implementation, similar to the existing Tensor_int8 and Tensor_float32.
• Update the dtype switches in Tensor.java to support bfloat16.
• Add corresponding tests to TestTensor.java.

CC #6571

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

cc @kirklandsign @cbilgin

[phaiting]

I'd like to work on this task. Can I have it assigned to me? I'll make sure to follow the guidelines and keep the conversation updated."

[kirklandsign]

Per discussion with @larryliu0820 @JacobSzwejbka we should not expose bf16 or fp16 as java input/output type.

Also per discussion with @phaiting seems this is not related to #6571

Let's close this one and fix #6571 instead?

[kirklandsign]

In https://github.com/pytorch/executorch/pull/10414/files#diff-f4be65c0ce84e4d84ea3e990e7ba0c6789800cf83a05bb7c37f278ade1339e19R703, we will use a unsupported subclass for bf16, fp16.