I found that there are some issues with the BackoffPolicy:

1. We should use a new backoff instance for each use:

   pulsar-client-go/pulsar/consumer_partition.go

   Line 1656 in 35ae051

   defaultBackoff = internal.DefaultBackoff{}

2. In the reconnection logic, the checking for IsMaxBackoffReached doesn't respect the pc.options.backoffPolicy.

   pulsar-client-go/pulsar/consumer_partition.go

   Line 1700 in 35ae051

   if maxRetry == 0 || defaultBackoff.IsMaxBackoffReached() {

   
   Actually, the IsMaxBackoffReached is incorrect. We should respect the operation timeout and check the whole retired time instead of the backoff delay time.
3. From the view of API design, we should move the BackoffPolicy out of the internal package.