Fix semaphore leak in VCS pool when jdbc connections are not available

Scenario to reproduce:

1. start DB in VM
2. start Nuxeo
3. try to access and perform operations on UI -> OK
4. pause DB
5. try to access and perform operations on UI -> KO
6. resume DB
7. try to access and perform operations on UI -> OK
8. stop DB in the VM but keep the VM up
9. try to access and perform operations on UI -> KO
10. start DB
11. try to access and perform operations on UI -> KO
12. this gives the following stack trace:

    2016-10-11 15:20:00,209 ERROR [Quartz_Worker-1] [org.nuxeo.ecm.core.event.impl.EventServiceImpl] Exception during triggerEsclationRules sync listener execution, continuing to run other listeners
    org.nuxeo.ecm.core.api.NuxeoException: Connection pool is fully used, consider increasing nuxeo.vcs.blocking-timeout-millis (currently 100) or nuxeo.vcs.max-pool-size (currently 20)
    	at org.nuxeo.ecm.core.storage.sql.ra.ConnectionFactoryImpl.getConnection(ConnectionFactoryImpl.java:108)
    	at org.nuxeo.ecm.core.storage.sql.ra.ConnectionFactoryImpl.getSession(ConnectionFactoryImpl.java:159)
    	at org.nuxeo.ecm.core.repository.RepositoryService.getSession(RepositoryService.java:230)
    	at org.nuxeo.ecm.core.api.local.LocalSession.createSession(LocalSession.java:117)
    	at org.nuxeo.ecm.core.api.local.LocalSession.connect(LocalSession.java:83)
    	at org.nuxeo.ecm.core.api.CoreInstance.acquireCoreSession(CoreInstance.java:184)