Only retry with retryable amqp errors for sender

[gavinfish]

Fix #200

1. When sender.trySend() fails with amqp error, it is not necessary to retry for all the conditions. Only retry with the below errors:

• amqp.Error
  • com.microsoft:server-busy
  • com.microsoft:timeout
  • com.microsoft:operation-cancelled
  • com.microsoft:container-close
• amqp.DetachError

You can refer to python sdk with similar logic at
https://github.com/Azure/azure-sdk-for-python/blob/c9240aa68c2870a1a72249b4c86f3ca95551cc29/sdk/servicebus/azure-servicebus/azure/servicebus/exceptions.py#L58-L80

2. Reduce the retry times from 10 to 3, delay from 10s to 1s. Take com.microsoft:server-busy with 10s delay which is suggested from service bus error.

"*Error{Condition: com.microsoft:server-busy, Description: The request was terminated because the namespace azuresbtests-fqfclayc is being throttled. Error code : 50009. Please wait 10 seconds and try again. Reference:3f49ca31-e885-4491-853d-ffaa8293c013, TrackingId:1a2c36240000000a0024729e5d39641b_G17_B2, SystemTracker:azuresbtests-fqfclayc:Queue:max-send-codvkdzlfb, Timestamp:2019-07-25T08:11:16, Info: map[]}

3. Do not close the sender when the error cannot be recovered. It is not necessary especially that the lib cannot handle it well with Topic struct well now. Better to add an option to determine whether close the sender in the future.

[catalinaperalta]

LGTM

Need to update changelog and add method comment.

[catalinaperalta]

Could you also add a test to mock receiving these errors and also update the changelog? I'll then approve again.

[gavinfish]

Could you also add a test to mock receiving these errors and also update the changelog? I'll then approve again.

@catalinaperalta I have some problems to add unit tests here. As amqp.Sender is a struct without interface, I cannot easily mock it to return different error messages here. I can also abstract functions to wrap calling amqp.Sender and mock the wrapper, but this is a little bit wired just for test.