* Improve `available` method of `LineOfferConsumer` class
* Add `quantity_available_for_offer` method to `AbstractLine` class
* Improve conditions: "Get num matches" methods updated to use `quantity_available_for_offer`
method of `Line` model.
* Improve conditions: `get_upsell_message` methods updated to show upsell message only in
the cases where `delta` > 0.
* Ensure basket upsell messages are not `None`

Fix handling of count conditions in `MultibuyDiscountBenefit`

* enforce max_affected_items during offer application

also restrict the usage of offer consumption to line.consume()

* improve test coverage

This change fixes the inconsistency with range.all_products() which
didn't check for exclusions when includes_all_products is set

Every other bit of oscar expected range.includes_all_products means
that all products are included, no questions asked. The dashboard
interface even hides the option to edit/remove individual products.

The short circuiting in ConditionalOffer.products() is removed so
the range.all_products() logic is applied consistently.

* fix bug in grouped_voucher_discounts

* add test for grouped_voucher_discounts

* remove 3.6.1 syntax

* improve test

- Refactor clean methods to return all validation errors, rather than just the first one we found.
- Fix clean method type errors in Python 3
- Add tests for the clean methods for all benefit types.

- If start_datetime is null, then the offer has no start date.
- If end_datetime is null, then the offer never expires.

Use the ActiveOfferManager in Applicator.get_site_offers instead of
reimplementing the same logic there.

Fixes #2344.

Now after ConditionalOffer deleted will be deleted related Benefits and Conditions models (through post_delete signal) if they are no longer used by other offers (with exception of custom benefits/conditions). Plus some small changes related to PEP8 and typos.

As in `test_model.py` -> `test_models.py`

The idea of splitting integration from unittests is good in theory
but leads to a lot of mental overhead. Besides whenever a tests
interacts with a database it isn't a unittest anyway.

Better to use filenames which are commonly used in the python word then
custom ones.

 - Set Meta.app_label on models created in the tests
 - Use logging.NullHandler instead of django.utils.log.NullHandler
 - Use oscar.core.loading.get_model() instead of django's get_model
 - Remove template tags '{% load url from future %}'

This should fix #1719.

Create a factory module per app and move them under the new
oscar.test.factories module.

Fixes #1709

The offer tests contained a stray test case that I suppose was once
intended as a base class for all the other tests. But it's only used
twice, and for very little gain, so let's delete it.

Django now throws a ValueError when an unsaved instance is assigned:
https://docs.djangoproject.com/en/1.8/releases/1.8/#assigning-unsaved-objects-to-relations-raises-an-error

This commit changes the offending lines to create objects in the
database instead. But I don't understand why I don't need to make
the same changes for e.g. the country factory.

But I don't really care, as long as the tests pass. I'm guessing it has
something to do with factory-boy and the get_or_create statement.

Unfortunately, this will make our test suite slower. But to make up for
it, Django 1.8 introduces the setUpTestCase method, which we can use
once Django 1.7 is removed.

This order offers the simplest API for common needs - as evidenced by the
fact that the apply_offers crook could be removed from two test files,
saving on imports as well.

the count of all_products. Tests included

This already worked only within site offers by way of default model
ordering. Site offers are probably the most common offers, however the
ordering of all offers has to be controlled by the applicator.

Conflicts:
	tests/integration/offer/applicator_tests.py

Cherry-picked from #1414.