Fix sparse_placeholder error when int is passed in shape argument

This fix tries to address the issue raised in #6749 where and error
```
ValueError: Tensor conversion requested dtype int64 for Tensor with dtype int32: 'Tensor("Const:0", shape=(2,), dtype=int32)'
```
is thrown out in case int is passed in the shape argument.

This fix fixes the issue and adds test cases for it.

This fix fixes #6749.

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