Handle AutoReconnect errors in find_one.

These aren't currently caught because the `collection.find_one` method
is called, and then it calls `find` without going through the proxy object.

Added tests for autoretry_read and MongoProxy.

Replaces Executable class with autoretry_read decorator.

The decorator can be used to wrap high-level methods so that
exceptions happening when iterating the cursor can be handled with
the same logic.

Also only specific read methods are retried.

Update packaging

Proxy executable methods from Database, ReplicaSetConnection

and mongo client classes

Set logger and wait time when creating nested MongoProxy object

Added Apache 2.0 license header

Added Apache 2.0 license header

Allow proxying of collections accessed by attribute
Attempt to reconnect for 60 seconds; make it configurable log delay time

The MongoDB documentation says it can take up to 60 seconds for a new primary to be elected when replicaset failover occurs (http://docs.mongodb.org/manual/faq/replica-sets/#how-long-does-replica-set-failover-take). Therefore, we need to wait up to 60 seconds for failover, not 16 seconds.

Make the wait interval configurable by adding an optional wait_time parameter to the MongoProxy initializer.

When logging attempts to auto-reconnect, include the elapsed time in the log message.

not just by __getitem__.

Injection logger

setup.py, robustness, user-friendliness

Only wrap things that are callable, so that, e.g., properties (like
"db.name") will continue to work.

Also, __getitem__ should fetch item from self.conn, and __getattr__
should fetch attr from self.conn, and they shouldn't be mixed up.