In 4.0 default database is `neo4j` and `graph.db`is the database in any
pre version. We will try to choose graph.db as a default database only
if neo4j does not exist yet. Before this PR that was not the case and
selection could happen any time.

Not adding any tests due to this only showing up when using a bolt
driver to connect the database and would need a driver with
multidatabase support to handle this case.

Temporary fix by increasing max size on allowed database names

The tests uses the unique test name as database name
which is up to 95 characters long
The error message is kept saying max 63 since that is the intention

Added better error messages and changed _ to -

Changed so assertValidDatabaseName() takes a DatabaseId instead of a
(assumed) normalized string

Moved NameValidator from the front-end

This by changing to DatabaseId instead of just String name

Added more validation to the database name

Both for standard default database, custom default database and system
database.

This will make the fix for not allowing to stop the default and system
databases easier. Also, since there are no other active databases in
community, the commands will no longer have any effect there.

In the big merge was some of the shuffling of tests
and the added tests therein missed, now added back

available exception in that case instead of failing with server error.

Always check that bolt transaction spi started for available database.
Bump driver to latest alpha version to be able to access different databases.
Make default temporal tx listener more failsafe.
Capture failed database context in case when database start failed.

Plan_with_minimum_graph replaces the previous functionality
plan_with_minimum_cardinality, and solves the issue of getting plans on
an empty database (or supersmall) that do not scale as data is added.
The previous solution would set a minimum bound for operator cardinalities,
but this would not work if the operator had an input cardinality bigger
that the specified min cardinality. The new solution simply sets
minimum values for the graph statistics, which should provide a
whole-planner emulation of a bigger than actual graph, and therefore
give scalable plans always.

 * Removed selectorThreads system property because we never head a need
   to configure it. Netty itself has the same system property to
   configure the number of threads. It calculates the default
   amount using the same formula

 * Create a single event loop group and allow Netty to use it for
   both parent (acceptor) and child (client) threads. This removes
   the need to create manage two event loop groups in Bolt server

 * Close all server channels before closing the event loop group.
   This seems to be the recommended way of closing down Netty
   servers. It is also an attempt to fix a rare event loop
   shutdown exception "eventfd_write() failed: Bad file descriptor"

 * Shutdown the event loop in a synchronous manner. This is to make
   sure all network resources are terminated when Bolt server stops

 * Remove explicit pooled allocator configuration because it is
   now the default

 * Deregister ports from the port register when Bolt server is stopped

Its useless for now and only causes failures of builds here and there
from time to time.

In 4.0, and Lucene 8, it no longer does this.
This changes reverts the standard analyzer to its former behaviour of filtering English stop words.

Replace custom Futures#failedFuture() with standard
JDK CompletableFuture#failedFuture().