[BUG] NodeBackup::spawn_replica does not release a reserved port number during failure

[dmitry-lipetsk]

Alter execution of TestTestgresCommon::test_replication_slots one port is not released.

Not released port is allocated in NodeBackup::spawn_primary that is called in NodeBackup::spawn_replica:

testgres/testgres/backup.py

Lines 187 to 192 in 5f8f5dd

	with clean_on_error(self.spawn_primary(name=name,
	destroy=destroy)) as node:
	# Assign it a master and a recovery file (private magic)
	node._assign_master(self.original_node)
	node._create_recovery_conf(username=self.username, slot=slot)

NodeBackup::spawn_replica uses clean_on_error to destroy newly created node but clean_on_error does not call PostgresNode::free_port.

testgres/testgres/utils.py

Lines 314 to 326 in 5f8f5dd

	@contextmanager
	def clean_on_error(node):
	"""
	Context manager to wrap PostgresNode and such.
	Calls cleanup() method when underlying code raises an exception.
	"""
	try:
	yield node
	except Exception:
	# TODO: should we wrap this in try-block?
	node.cleanup()
	raise

It seems to me, to fix this issue we can just call node.free_port method in clean_on_error immediatelly after node.cleanup.

Or we must to call node.free_port from node.cleanup.

[dmitry-lipetsk]

It is a part of work for #247.