Initial pubsub implementation plan

[tseaver]

• Sketch out "science-fiction" doctests showing "normal path" API usage, and get review from back-end. (Sketch 'gcloud.pubsub' API. #691)
• Implement pubsub.connection.Connection class (based on storage version). (Flesh out pubsub topics #742)
• Implement pubsub.topic.Topic class and related search API. (Flesh out pubsub topics #742)
• Implement pubsub.subscription.Subscription class and related search API. (Flesh out pubsub subscriptions #743)
• Fold in implicit project / connection stuff, shared with storage. (per Unify '_implicit_environ' handling #741, see Add implicit environment support to pubsub #763)
• Update pubsub.api.list_topics() to return pubsub.topic.Topic instances (#744: return Topic instances from pubsub.api.list_topics() #790)
• Update pubsub.api.list_subscriptions() to return pubsub.subscription.Subscription instances (#744: return Subscription instances from pubsub.list_subscriptions() #794)
• Update pubsub.subscription.pull() to return instances of a new pubsub.message.Message class (e.g., to handle ack_id, id, payload decoding, maybe other behavior). (#744: Add 'pubsub.message.Message' class #800)
• Add (pull-only) regression test suite. (Begin pubsub regression suite #765. #744: Add E2E regression test for message publication + pull #795)