Nothing should be transmitted when STARTTLS negotiation fails, not even
a stream close.

When STARTTLS fails, there are two distinct types of failure. Each type should
be handled in a distinct manner.

Note that this commit applies a fix to the S2S code only. C2S may also need work.

Fix NPE when accessing AuditManager module

Not sure when this got changed, but the module loader uses the interface
class name as a key, not the implementation class name.

Avoid using classes from sealed packages

This fix was applied a while back on an earlier incarnation of this test
class.

OF-122: Fix inconsistent MUC subject handling

OF-993: Replace thread factory boilerplate code.

Ensure that the MUC room subject is set correctly from initial request
and upon subsequent reload.

Subscribe with Response Unsubscribed Causes Roster Push to Responding Client

Assume a client with MUC elements in presence needs to join

OF-992: Re-init on restart

OF-991: Ant build should use property values when available.

Fix XEP-0012 logic.

JDBCAuthProvider: adding support for bcrypt and more

Prevent WritePendingException during delivery of BOSH packets.

If a query has no 'to' attribute it should be treated on behalf of the sender.
It should not be treated as if it were targeted to the server (and return the server uptime).

The former logic also threw an UserNotFoundException, if the user queried itself, because the user was not on his own roster.
 I've fixed that in the canProbePresence method.

 Fixes issue OF-982

This exception caused serious trouble, e.g.:
- Clients got an invalid/incomplete packet, which breaks them.
- Openfire logged related exceptions like org.jivesoftware.openfire.http.HttpConnectionClosedException: The http connection is no longer available to deliver content

It was caused by calling complete() although the previous write() has not completed (due to its async nature).

I also restructured that the Content-Length header is only set once.

Fixes OF-989.
See also https://community.igniterealtime.org/thread/57622

OF-983 Do not bounce packets while lock is held

OF-984 / OF-985 Dialback fixes for to attribute and TLS

OF-631 Reduce salt size to fit database

The salt size was 32, and the database column was a VARCHAR(32).

However, the salt was base64 encoded prior to insertion, and this caused it to
be too large. Therefore this patch reduces the size to 24 bytes, which expands
to 8 units of three bytes encoded as four characters, hence 32 characters total.

OF-631 Fix all the SCRAM faults

A few places all use the domain.intern() as the key for a synchronized block.

This can lead to clashes when multiple semantics conflict, and this can in turn
lead to a deadlock, as witnessed in OF-984.

This changes two cases to use a prefixed string to namespace the locking.

OF-983 appears to be caused by a complex bouncing process run while the
authentication check lock is held. This patch moves the actual bounce to
outside the synchronized block, so the code no longer deadlocks.

Untested, but based on the stack traces in OF-983 reports.

Revert "OF-115 Remove race during outgoing session creation"

Ignore test that makes Travis fail.

This reverts commit d0667e21.

While I think the general code is right, it's clearly got serious shortcomings
so I think it's best to back this out for now.

Refactor connectionmanagement