Matching engine fix for flaky unit test (cadence-workflow#349)

TaskListManager needs range to updated before starting task writer so it
can initialize max read level correctly using the initialized
taskSequenceNumber on TasklistMgr.  If ordering of reversed then
MaxReadLevel is initialized incorrectly to -1 and will rely on a task to
be written before it can be set to correct value.  This results in
GetTasks to use incorrectly initialized MaxReadLevel and will never read
tasks from persistence store.

service/matching/matchingEngine_test.go

@@ -827,7 +827,7 @@ func (s *matchingEngineSuite) TestMultipleEnginesActivitiesRangeStealing() {
 	workflowExecution := workflow.WorkflowExecution{RunId: &runID, WorkflowId: &workflowID}
 
 	const engineCount = 2
-	const taskCount = 200
+	const taskCount = 400
 	const iterations = 2
 	const initialRangeID = 0
 	const rangeSize = 10
@@ -985,7 +985,7 @@ func (s *matchingEngineSuite) TestMultipleEnginesDecisionsRangeStealing() {
 	workflowExecution := workflow.WorkflowExecution{RunId: &runID, WorkflowId: &workflowID}
 
 	const engineCount = 2
-	const taskCount = 200
+	const taskCount = 400
 	const iterations = 2
 	const initialRangeID = 0
 	const rangeSize = 10

service/matching/taskListManager.go

@@ -128,14 +128,17 @@ type syncMatchResponse struct {
 func (c *taskListManagerImpl) Start() error {
 	defer c.startWG.Done()
 
-	c.taskWriter.Start()
-	c.signalNewTask()
-	go c.getTasksPump()
+	// Make sure to grab the range first before starting task writer, as it needs the range to initialize maxReadLevel
 	err := c.updateRangeIfNeeded() // Grabs a new range and updates read and ackLevels
 	if err != nil {
 		c.Stop()
 		return err
 	}
+
+	c.taskWriter.Start()
+	c.signalNewTask()
+	go c.getTasksPump()
+
 	return nil
 }