Allow _FilterFloat to pass None values. (#13834)

* Allow _FilterFloat to pass None values.

[Tensorflow 1.3: tf.constant with dtype=[float32, float64, float16] may have inconsistent behavior. #13827](https://github.com/tensorflow/tensorflow/issues/13827)

 - This fix allows _FilterFloat to allow None values to pass.
 - We add `tf.float16` into the `_TF_TO_IS_OK` dictionary.
 - We add tests for `convert_to_tensor` and `tensor_util`.

* Added `tf.float16` into the _TF_TO_IS_OK dictionary

[Tensorflow 1.3: tf.constant with dtype=[float32, float64, float16] may have inconsistent behavior. #13827](https://github.com/tensorflow/tensorflow/issues/13827)

This fix adds `tf.float16` into the `_TF_TO_IS_OK` dictionary for consistent error checking.
A note is added to `convert_to_tensor` on how `None` values are handled.