get_serializer_class doesn't recreate same conditions as would be used by DRF

Nobody has claimed this yet.

Where `get_serializer_class()` is used to determine which serializer would be used for a `ModelViewSet`, it doesn't re-create the same environment as DRF would use when servicing an actual API request, and so it may return different results (and thus incorrect documentation).

As an example, I have the following method, which uses a cut-down serializer for list actions, and a full-content serializer for retrieve actions:

```python
class FooViewSet(viewsets.ReadOnlyModelViewSet):
queryset = Foo.objects.all()

def get_serializer_class(self):
if hasattr(self, 'action'):
if self.action == 'list':
return FooListSerializer
elif self.action == 'retrieve':
return FooSerializer

return serializers.Default
```

drfdocs doesn't currently set the `self.action` member before calling `get_serializer_class()`, so it gets the default serializer with no fields, whereas an actual API request would get either the `FooSerializer` or the `FooListSerializer`.

Start from the get_serializer_class() call used during documentation generation and compare its setup with the ModelViewSet conditions described in the issue. Ensure the action context matches list and retrieve API requests, then verify that generated documentation uses FooListSerializer and FooSerializer rather than the default serializer.

Written by the indexing model from the issue text.