`retry_exponential_backoff` algorithm does not account for case when `retry_delay` is zero

[edwardwang888]

Apache Airflow version: 2.1.2

Kubernetes version (if you are using kubernetes) (use kubectl version):

Environment:

• Cloud provider or hardware configuration:
• OS (e.g. from /etc/os-release):
• Kernel (e.g. uname -a):
• Install tools:
• Others:

What happened:
When retry_exponential_backoff is enabled and retry_interval is inadvertently set to zero, a divide by zero error occurs in the modded_hash calculation of the exponential backoff algorithm, causing the scheduler to crash.

What you expected to happen:
Scheduler should treat it as a task with retry_delay of zero.

How to reproduce it:
Create a task with retry_delay=timedelta() and retry_exponential_backoff=True.

Anything else we need to know:
Willing to submit a PR; opened #17003 (WIP) with possible fix.