Use correct server's start time

When a machine is resized, rescued or evacuated the server's start_time
(and the usages' start_time) is reset and it does not reflect anymore
the time the instance was created. Instead of relying on that time, we
move to use the server.created time.

This solves the issue of negative accounting times, that happen when an
instance has its start_time reset.

Fixes #56

caso/extract/nova.py

@@ -198,7 +198,7 @@ def extract_for_project(self, project, extract_from, extract_to):
             records[instance_id].disk = usage["local_gb"]
 
             # Start time must be the time when the machine was created
-            started = dateutil.parser.parse(usage["started_at"])
+            started = server_start
             records[instance_id].start_time = int(started.strftime("%s"))
 
             # End time must ben the time when the machine was ended, but it may