Decide how to handle stream vs bytes content in Speech streaming_recognize().

See: https://github.com/GoogleCloudPlatform/google-cloud-python/pull/2680#discussion_r86837992

`Sample.content` needs to be a file-like object when calling `streaming_recognize()`. Right now `ValueError` is raised if it is not. But there may be some other, more clear implementation.