Ha mode switcher defined in ha module and placed in life
before some of the database components. Its possible that voting and
even switching to a new state will happen before all of the components of
platform life will gonna finish its start.
To be sure that we stopping services (on slave switch)
that are already started and we do not race anywhere
we will gonna wait till database will be available before doing that.

This reverts commit ad3b3e447cf4087b7d2980f5e03872f27a464a7b.

This reverts commit fee00505d1847d9a63ec385fe73f179e8c2b574c.

# Conflicts:
#	public/enterprise/cypher/acceptance-spec-suite/src/test/scala/cypher/features/BaseAcceptanceTest.scala
#	public/enterprise/cypher/compatibility-spec-suite/src/test/scala/cypher/features/BaseTCKTests.scala
#	public/enterprise/cypher/spec-suite-tools/src/test/scala/cypher/features/BaseFeatureTest.scala

This way, we can distribute the tests across multiple parallel runing VMs, utilize more CPU power and effectively reduce waiting time

Introduce random extension.

A lot of stress tests are failing because these statuses are not
considered transient. This should solve that issue.

JDK 9+ does not contain JAXB. It can be added with
"--add-modules java.xml.bind" for JDK 9 and 10. However, JAXB is
completely removed from JDK 11. This commit adds a new profile to the
neo4j-server module  with additional dependencies on JAXB to make
`ServerConfigIT` pass.

It is added by a Maven profile which was previously only activated on
JDK 9. This commit makes it activate on both JDK 9 and 10. Absence
of java.xml.bind made `ServerConfigIT` fail.

Previously, decoding of inbound Bolt messages and their routing to the
state machine was very protocol V1 specific and not extensible. This
commit makes it possible to register protocol version specific message
deserializers. It also makes code always use same inbound message
classes. Two different implementations existed before - one for
production code and one for tests.

Central component is `BoltRequestMessageReader` which holds a
configurable set of message deserializers. Correct deserializer
is looked up by a message signature.

switch.

Switch to usage of ha module directly since dependency resolver is not always
available to provide role and and state information.
(For example case where data source was not started yet or restarting now )

This commit fixes a problem where configured time zone for log files
was not used for the user log. `Neo4jRule` used to ignore the
`dbms.db.timezone` setting. Thus neo4j.log and debug.log contained
different time zone offsets in logged timestamps.