Fix issue of `name` used in `build_default_serving_input_fn` (#12759)

* Fix issue of `name` used in build_default_serving_input_fn

This fix tries to address the issue raised in 12755 where the
name used in `build_default_serving_input_fn` is not a valid
variable scope:
```
$ python -c "import tensorflow as tf
f = {'feature': tf.placeholder(name='feature', shape=[32], dtype=tf.float32) }
serving_input = tf.contrib.learn.utils.input_fn_utils.build_default_serving_input_fn(f)
serving_input()"
...
ValueError: 'feature:0' is not a valid scope name
```

In `build_default_serving_input_fn`, the name of the tensor
is used directly which consists of the op name and the value index:
```
  @property
  def name(self):
    """The string name of this tensor."""
    if not self._op.name:
      raise ValueError("Operation was not named: %s" % self._op)
    return "%s:%d" % (self._op.name, self._value_index)
```

This fix fixes this issue by using the name of the tensor's op (not tensor):
`t.op.name` instead, to avoid the ":0" (value index) at the end.

This fix fixes 12755.

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

* Add test case for 'name' in build_default_serving_input_fn

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

* Add test case for build_raw_serving_input_receiver_fn

Also use t.op.name instead of split the name with ':'

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