Cleaner as_pymongo

[wojcikstefan]

Fixes #563 and closes #1002.

This PR introduces tweaks to QuerySet.as_pymongo - a method that hasn't been touched in at least a couple years. Part of me wants to get rid of it entirely and suggest that people use Doc._get_collection().find(...) if they need raw PyMongo documents, but perhaps that's too drastic of a change (and can be revisited when we make _get_collection public, see #1182).

This PR significantly simplifies the code and gets rid of an arguably useless coerce_types kwarg (the idea of converting values to MongoEngine-specific types in a method that's supposed to return raw PyMongo documents confuses me greatly). For now I also followed the previous logic of excluding _cls as an implementation detail and also excluding _id if it was only fetched for internal MongoEngine purposes.

Even though I suspect this to be barely used, it introduces a breaking change and should hence be released along with a major version bump.

---

This is off-topic, but I was curious and performed a test of how much faster as_pymongo is compared to the regular queryset iteration:

In [12]: class User(Document):
    ...:     name = StringField()
    ...:     email = EmailField()
    ...:

In [13]: for i in range(100000):
    ...:     User.objects.create(name='Steve ' + str(i), email='steve%d@gmail.com' % i)
    ...:

In [14]: def test():
    ...:     for doc in User.objects:
    ...:         pass
    ...:

In [15]: def test2():
    ...:     for doc in User.objects.as_pymongo():
    ...:         pass
    ...:

In [16]: %time test()
CPU times: user 8.75 s, sys: 179 ms, total: 8.93 s
Wall time: 8.89 s

In [17]: %time test2()
CPU times: user 463 ms, sys: 37.8 ms, total: 500 ms
Wall time: 517 ms

Quite a difference...