Inconsistency on QueryDict.items and documentation

[Daniel F Moisset]

There is a multiway inconsistency about the defined behaviour of QueryDict.items (a similar thing happens for QueryDict.values`:

• The ​current implementation actually returns a generator object (supporting iteration only)
• The ​documentation says that it is "just like the standard dictionary items() method", which in Python3 returns a dict_items object (a ​view object, supporting set-like operations)
• But the example in the documentation 2 lines below that show a standard python list (supporting indexing)

I see 2 different was to fix this:

1. Update the documentation and make explicit that a generator is returned and that Querydict is a bit different than python dicts
2. Make QueryDict behave more like python by returning a set(). This should not break backwards compatibility (a set is more general than a generator), but use more memory (because the set must be fully built instead of generated lazily)

I can provide patches for any of these options, but some core devs recommended me getting an opinion of which is the right solution before moving forward.