Fix dependencies overhead

[slafs]

This is the initial approach of fixing #1728 i.e. resolve the dependencies overhead and allow for installing minimal set of dependencies for a given service (e.g. when using only s3 backend, install dependencies used by moto.s3 module only).

Currently this is backwards incompatible (!!!)
and probably fails on some tests.

Before I start any efforts on backwards compat, I'd like to get some guidance on how to progress with this, first.

In order to reduce required dependencies we need to change the way we're accessing mock decorators (in moto/__init__.py) and the way we import backends (in moto/backends.py).
This might need an API change, so instead of doing from moto import mock_s3 we'd need to go for from moto.s3 import mock_s3.

BTW This PR reduces the time needed for import moto from 3.5 seconds to 0.5s (on my machine).

Things to discuss/decide:

• is this approach even feasible for @spulec
• should we use the new _backends variable or how we should avoid using it?
• do we still need botocore.awsrequest and ConnectionCls fix in moto/__init__.py (why do we need it anyway)?
• do we want to include ALL services in extras_require (in setup.py)?
• can we trim down more dependencies (i.e. remove any from install_requires still, e.g. werkzeug or Jinja2)?

[coveralls]

Coverage decreased (-0.2%) to 92.964% when pulling f710c7b on slafs:moto-1728/dependencies-overhead into e4a4e61 on spulec:master.

[edisongustavo]

Hello @slafs , if there is something I can do to help get this PR merged, I'm happy to help.

[edisongustavo]

This might need an API change, so instead of doing from moto import mock_s3 we'd need to go for from moto.s3 import mock_s3.

What about a potential workaround to this be, on moto/__init__.py, to avoid breaking the API:

def mock_s3():
    from . import s3
    return s3.mock_s3()

[slafs]

Hello @slafs , if there is something I can do to help get this PR merged, I'm happy to help.

Hey @edisongustavo. Thanks for the offer. At this point I think I'd like some input from someone responsible for this codebase, so we get at least an initial acceptance for this work. I guess I'd like some help on keeping this PR fresh (i.e. rebasing on top of spulec:master).

What about a potential workaround to this be, on moto/init.py, to avoid breaking the API:

Good idea! I'll try to incorporate this here. Then it should be easier to keep the PR up-to-date as well.

[slafs]

Closing in favor of #2697 and #2676