As a quick experiment, I tried out #29263 to not wait on the ByteStream being cleared out before we invoke recv_trailing_metadata and that started causing issues in our server streaming tests which expect the messages to be received before the status.

I think we should be able to fix it if we change our surface to always wait for recv_message ops to complete and the byte stream to be read before completing the recv_status op, but this change seems easier.

Thanks for reviewing!