Update docs for tf.cast with supported types

This fix tries to address the issue raised in 18529 where
there are some confusion over supported types for `tf.cast`.
This fix updates the docs with explicitly supported numeric
type for `tf.cast`:
```
`uint8`, `int8`, `uint16`, `int16`, `int32`, `int64`,
`float16`, `float32`, `float64`, `complex64`, `complex128`, `bfloat16`.
```

This fix fixes 18529.

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