When the connection to the primary master node is lost, the node manager
does not have anymore a reliable list of running nodes, so iterateForObject()
must not retry any cell.

Since transactions have metadata like a description, it may not be useless
to allow them. But the behaviour of FileStorage is to silently drop them,
so we may have to do the same in the future.

An application that is not supposed to commit empty transactions should write
its own unit test to prevent this.

This happened between storage nodes of different clusters because they're not
informed about their state, e.g. a dead upstream storage node.

In any case, logs were flooded at 100% cpu usage.

Binding a port actually does not reserve it. And on the other side, a bound
socket can't be bound again. So it could bind a port twice, warn about this,
and then raise EINVAL when trying to bind again.

Apart from this, the global lock didn't even prevent conflict with another
NEO test run when tests restart nodes. So better keep it simple.

Limiting the size of data.value column to 16 MB saves 1 byte by switching
to MEDIUMBLOB, and it avoid the need of big redo logs in InnoDB.

This fixes the case of an application that would store 2 values X & Y
where NEO internally compresses X into a value identical to Y.

This speeds up 2 operations that were horribly slow (linear scan
of source database, from the beginning or from the end):
- read access to not yet migrated data
- resumption (restart of storage node whereas import was not finshed)

Testing will tell if this patch is enough, or if more efficient solutions are
required, like:
- building a preliminary {tid->file_pos} index while the source DB is still
  in use (in this case, NEO may have to deal with a small gap at the end)
- disabling ZODB features requiring data_serial (undo)

Before it was only possible to specify them in a configuration file.

Value must be cast as integer.

The whole configuration part should be reviewed to use argparse
and avoid duplicated code.

They're decompressed automatically to temporary files.

The use of ZODB.broken was completely wrong so previous code only worked
if NEO could import all classes used by the application.

RTMIN+1 & RTMIN+2 signals, which were previously used for debugging,
have been remapped to RTMIN+2 & RTMIN+3

No journal on disk anymore because this is incompatible with the
rename+reopen way to rotate logs, and we want to support logrotate.
Of course in case of crash the log may be corrupt or entries may be lost,
but we currenty don't think NEO logs would be useful to debug such failure.

This commit only moves code from setup() to _setup()

The only purpose of committing during import is to not restart from the
beginning in case of interruption.

If not implemented, this should not result in a fatal error that kills the
storage. Instead of that, a new error packet is sent so that the exception is
raised in the client.