Plus reorganise tests

Validation is based on wikipedia list of postal codes [1]. Formats were
converted into regexes for countries.

1: http://en.wikipedia.org/wiki/List_of_postal_codes

Fixes #753

Fixes #745

Bad index in `last_line = covered_lines[-1][1]` (previously [-1][0])
caused rounding error. When you have 3 products for £7 and benefit 3
things per £20, each product would get discount £0.33 which doesn't sum
up to £1. The existing code handles this but bad indexing makes it fail.
A test case reproducing the issue was added.

Fixes #742

This change adds a uniqueness constraint on user addresses to prevent
duplicates.  A few changes follow:

- The user address forms in account and checkout are modified to take a
  user as a constructor arg so uniqueness can be checked.

- The assignment of a user when creating a new address has been moved
  into the form.

Also I moved the generate_hash fn up to the Address superclass as it's
possible that it might be used in other Address classes.

If parent is given, only parent descendant categories will be returned.

Fixes #741

- Add more links
- Add event details to line detail page

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.

* Tidy implementation of validator
* Tests improved and moved to tests/integration (as they call resolve
  internally)

As they save stuff...

Without a value, applying an AbsoluteBenefit raises a TypeError.

Fixes #670

The auth backend now gracefully handles situations where there are users
with the same email address.

Introduced in 7fd62f2af0.

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.

They aren't treated specially any more.  They become part of the
'applications' collection rather than being attached to the basket.

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

...as they are all integration tests (they save stuff).

This uses the unidecode package to convert non-ASCII words to ASCII
equivalents.  We use a new slugify function to wrap this logic as we
will want to use it elsewhere.

See this StackOverflow thread for context:
http://stackoverflow.com/questions/702337/how-to-make-django-slugify-work-properly-with-unicode-strings/

Unidecode becomes a new dependency for Oscar.
https://pypi.python.org/pypi/Unidecode

Hat-tip to @avsazonov for suggesting this fix.

New impl is considerably simpler and lets the save method update each
node's slug and full name.

They use the DB so they are not really unit tests.

This makes it more efficient to load offers now that we can't simply use
the date filters to exlucde unavailable offers.

The is_active field has been reworked to depend on the
get_max_applications method.

The save method of ConditionalOffer now auto-sets status.

Fixes #403

This might be futile.

So we add handling to the filter where we check the value can be
converted to a valid decimal.  If not, return an empty string.

Fixes #316

Plus fix a bug when handling None.