* enforce max_affected_items during offer application

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

* improve test coverage

- 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.

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

Cherry-picked from #1414.

Testing a query raised in this thread:
https://groups.google.com/forum/?utm_medium=email&utm_source=footer#!msg/django-oscar/vceEpy889O4/OBzt6p9rsywJ

This involves some quite involved changes to the internals of how offers
and organised.  A new boolean field has been added to the offer model to
indicate whether to apply to prices exclusive of tax.  This is used
within the offer methods to determine whether a basket meets a
condition, and how to apply the benefit if it does.

Fixes #925
Fixes #924

Hmmm, this turned out to be a monster commit.  This change allows the
basket to be able to correctly calculate prices including tax.

It also requires a whole load of test changes since all baskets now
require a strategy instance to be assigned.

This allows updating code in both places, if necessary. Getting Django
1.5 support would be painful otherwise. This also removes the circular
dependency of django-oscar-testsupport on Oscar.

The requirements were merged. The imports were updated
accordingly. Unused imports in the touched files were removed. No
further changes.

Without a value, applying an AbsoluteBenefit raises a TypeError.

Fixes #670

This change applies the discount to all matching lines in the benefit
range, weighting the discount by the line prices.  The old way was to
iterate over each line and try and apply all the discount.

This new way means that applying an absolute benefit consumes more
products from the basket than the old way, which may possibly have some
confusing side-effects.

This change avoids an issue where customers can get a free product if
they return most of the ordered products apart from the one that was
discounted to zero.

This paves the way for custom benefits to return successful applications
that don't lead to a discount.

This might be futile.

Fixes #376

This is a major clean-up of offers, tidying up the implementations and
rewriting the test suite.  This should also clean up some edge cases
around the use of max_affected_items.