nack backoff policy may not be used

[davendu]

In PIP-106 and #660, the Pulsar SDK for Golang has supported NackBackoffPolicy. However, using a topic with 1 partition, the following code may not use the backoff policy:

polarisConsumer := p.consumers[priority], err = client.Subscribe(pulsar.ConsumerOptions{
	Topic:            "demoTopic",
	SubscriptionName: "demoSubscriptionName",
	Type:             pulsar.Shared,
	RetryEnable:      true,
	DLQ: &pulsar.DLQPolicy{
		MaxDeliveries: 10, 
	},
	EnableDefaultNackBackoffPolicy: true,
})
msg := <-polarisConsumer.Chan()
msg.Nack(msg)

After some digging I found the following call stack:

• Nack for a message is implemented in pulsar/consumer_impl.go, it may call the consumer carried by a trackingMessageID
• trackingMessageID defined Nack() at pulsar/impl_message.go, calling NackID()
• NackID() is implemented at pulsar/consumer_partition.go, calling Add() method for the nack tracker.

And in the implementation of Add() method, it's obviously impossible to use the back off policy, since redeliver count is not available, so default back off delay is used.

I'm not sure if my code is using the back off policy in the right way, or this is a bug in the SDK, or our broker is having some issue.

[flowchartsman]

Isn't this fixed by #974?