InputObjectType.__init_sublcass_with_meta__ skips Meta.fields attribute

In `InputObjectType.__init_subclass_with_meta__`, the `fields` of the `Meta` class (passed as an `options` arg) are never referenced. This in effect just skips them completely wich stands in contradiction to the docstring.

```
Meta class options (optional):
...
fields (Dict[str, graphene.InputField]): Dictionary of field name to InputField. Not
recommended to use (prefer class attributes).
```

`InputObjectType` has to be subclassed and `__init_subclass_with_meta__` has to be extended for this feature to work, like in [graphene.tests.issues.test_720](https://github.com/graphql-python/graphene/blob/efe4b89015d56a0b6c81abc6c0b4e46984aefd11/graphene/tests/issues/test_720.py#L9).

If this is accepted I can do a PR for this along the lines:
```diff
@classmethod
+ def __init_subclass_with_meta__(cls, container=None, _meta=None, fields=None **options):
- def __init_subclass_with_meta__(cls, container=None, _meta=None, **options):
if not _meta:
_meta = InputObjectTypeOptions(cls)

+ if not fields:
+ fields = {}
- fields = {}
for base in reversed(cls.__mro__):
fields.update(yank_fields_from_attrs(base.__dict__, _as=InputField))

if _meta.fields:
_meta.fields.update(fields)
else:
_meta.fields = fields
if container is None:
container = type(cls.__name__, (InputObjectTypeContainer, cls), {})
_meta.container = container
super(InputObjectType, cls).__init_subclass_with_meta__(_meta=_meta, **options)
```

Start at InputObjectType.__init_subclass_with_meta__ and compare its handling with the documented Meta.fields option. Review graphene/tests/issues/test_720.py for the referenced behavior. Done means Meta.fields is preserved and covered by a test demonstrating that subclassing InputObjectType does not discard those fields.

Written by the indexing model from the issue text.