Netty upgrade

Remove mistakenly added license header

Fix #4829: Behaviour of Path.lastRelationship.

This reverts the behaviour of Path.lastRelationship, which was
accidentially changed to return the first, not the last relationship
during a refactoring: 92ab0c60

1.9-maint version

1.9 fix gremlin build

This POM doesn't represent a published module. It's just used as a
convenience for invoking a third-party tool. Because it's not hooked
into the Neo4j parent module, updating the version number is a special
case that requires a workaround. Therefore we have decided to keep the
version number fixed and never change it. This change updates the
version number to something that clearly shows it's not related to the
Neo4j version number.

Fixes lifecycle issue after multiple starting/stopping dbs

Issue was that starting/stopping a database multiple times in the same
neo4j desktop application instance had each logging line occur as number
of times as the database had been started. This was due to the same
LifeSupport instance being used, and an additional Logging instance added
to it every time the database started.

Solution is to instantiate a new LifeSupport for every started db.

Fixes a race condition in HighAvailabilityModeSwitcher

Fixes a race between a SwitchToSlave rerunning because of
 either a NoSuchLogVersion or MismatchingStore Exception
 and a masterIsElected for the same or different master. That
 could lead to the availableMasterId being set to null and the
 second run of the SwitchToMaster failing with a NPE.

DumpStore with list of ids.

Syntax like: DumpStore file:id1,id2

due to builds hanging sometimes when using the sub-process way of crashing
a database. Also asserts highId after a successful recovery.

Updating license header

Fixed invalid epoch exception in ClusterTopologyChangesIT

More detailed error message when id generator capacity is exceeded

Previously the test slaveShouldServeTxsAfterMasterLostQuorumWentToPendingAndThenQuorumWasRestored might throw InvalidEpochException at node creation after epoch fix. The error arises because we do not ensure that the node creation definitely happens after epoch fix. This pr adds a listener to ensure that node creation should always happen after epoch fix.
Also we refined logging info in HighAvailabilityMemberStateMachine

Makes Learner give up properly when a value is impossible to learn

Cluster members previously would keep retrying to learn a Paxos
 instance they had missed even if all other cluster members already
 had responded with learnFailed. These retries would happen without
 any rate limiting and by definition they would never complete. One
 symptom of this bug was the filling up of messages.log with
 LearnFailure messages.
This patch removes that erroneous behaviour by simply logging
 the failure to learn on the delayed learner (and not on the
 properly up-to-date instances as before) instead of doing
 any retries. That is good enough because an instance when it
 realizes it misses a value it asks all live cluster members
 at once.
This patch also adds a bunch of tests on LearnerState to ensure
 proper responses on learn requests and learnFailed responses.