Fix deadlock with storages that "sync" on a new transaction

This backports a change from commit 227953b977a9e195c4ce9bbb9acd9c5ee60c333a.

NEO, as well as ZEO+server_sync (ERP5 backports this feature with a
monkey-patch), pings the server (primary master node in the case of NEO) on
new transactions. However, this round-trip is actually performed by the thread
that also does tasks requiring to lock the DB, like processing of invalidations.

Since transaction 1.6.1 (more precisely commit e581a120a6), IStorage.sync()
is called indirectly by DB.open() when a transaction has already begun,
and the DB must not be locked when this happens.

src/ZODB/DB.py

@@ -753,19 +753,16 @@ class DB(object):
                     result = self.pool.pop()
             assert result is not None
 
-            # open the connection.
-            result.open(transaction_manager)
-
             # A good time to do some cache cleanup.
             # (note we already have the lock)
             self.pool.availableGC()
             self.historical_pool.availableGC()
-
-            return result
-
         finally:
             self._r()
 
+        result.open(transaction_manager)
+        return result
+
     def connectionDebugInfo(self):
         result = []
         t = time.time()

src/ZODB/tests/testDB.py

@@ -151,7 +151,7 @@ def connectionDebugInfo():
     >>> before
     [None, '\x03zY\xd8\xc0m9\xdd', None]
     >>> opened
-    ['2008-12-04T20:40:44Z (1.40s)', '2008-12-04T20:40:45Z (0.30s)', None]
+    ['2008-12-04T20:40:44Z (1.30s)', '2008-12-04T20:40:46Z (0.10s)', None]
     >>> infos
     ['test info (2)', ' (0)', ' (0)']