Update documentation for uint16 support in `tf.resize_...` ops (#13721)

* Update documentation for uint16 support in `tf.resize_...` ops

This fix tries to address the different between the registered
kernels and the documentation differnet for uint16 support of:
`tf.resize_area`
`tf.resize_bicubic`
`tf.resize_bilinear`
`tf.resize_nearest_neighbor`

Signed-off-by: Yong Tang <yong.tang.github@outlook.com>

* Add test cases for uint16 of `tf.resize_...`

Signed-off-by: Yong Tang <yong.tang.github@outlook.com>

* Add test case of float16 for `tf.resize...`.

Signed-off-by: Yong Tang <yong.tang.github@outlook.com>