This adds addRelation and removeRelation capabilities to Role, making it possible to add users to the users column and roles to the roles column in a Role object, for example. This prevents the error of Role not having the attribute addRelation or removeRelation when trying to add users or roles to a Role, which is critical for Role functionality.

Update datatypes.py to handle dict on properties

As a dict's iterator returns keys, sending {"key" : "value"} resulted in receiving ["key"] on dashboard
Uses dict.iteritems() instead

add file support

Reverting 2fde6572 to fix #67 (lazy loading fails)

Added role and ACL support

Improve error messages.

Subclassing User object

fixed python 3 issue with str having __iter__ attribute

ParseResource's need to know they've been loaded

In __getattr__ we check to see if the object has been lazy loaded or not
before we try to do attribute lookups. However, once the attributes have
been loaded, we never record that.

So this changes _init_attrs, which actually does the loading, to record
that it's done so.

This change took a loop I had that was taking 8-10s to finish down to
.5-1s.

Fix the relatedTo Query operator

Basically, if we're asked to convert_to_parse a list, do it for
everything in the list and return the list.

All of the tests pass now, so I don't believe anything has been harmed
in this change.

It currently fails because:
  TypeError: <Game:XKz5Hfyxqw> is not JSON serializable

Before this we were never passing in all of the data required for the
relation lookup to work.

Figured this out after looking at the 4th code example at
https://parse.com/docs/rest#queries-relational and comparing to the
request that was actually being sent.

All the tests pass now.

Making an Object, and then marking that it's got Relation's doesn't
work. These tests demonstrate that. Next commit will fix whatever's
needed so that these tests pass.

great, thanks

Add timeout to urlopen

- Connection can sometimes hang forever on SSL handshake

eg Subscription.Query.get(user=User(objectId='foo'))

Python3, tests for new features, small fix for __len__ and query copying

Object.factory returns user defined types if exists. Also returns same type for two calls with same arguments.

independent derived queries - filter, limit, skip, select_related methods returns queryset copy instead of self

This is proper implementation of idea suggested by orignal limit and skip methods based on metaclass (but not working at all)
It works same as Django queries - original unfiltered query can be still executed or used to create
different filter.

made skip and limit consistent with other query methods