ContainerMetadataCredentials should retry with backoff

[bennettgoble]

When running aiodynamo in an ECS application under high load it is possible to hit timeout issues accessing the container metadata endpoint. I believe the chance of re-connection could be improved if the retry logic implemented some basic backoff.

Traceback (most recent call last):
  File "uvloop/cbhandles.pyx", line 63, in uvloop.loop.Handle._run
  File "/opt/venv/lib/python3.12/site-packages/aiodynamo/credentials.py", line 289, in _set_current
    self._current = task.result()
                    ^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/aiodynamo/credentials.py", line 389, in fetch_metadata
    response = await fetch_with_retry_and_timeout(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/aiodynamo/credentials.py", line 604, in fetch_with_retry_and_timeout
    raise TooManyRetries()
aiodynamo.credentials.TooManyRetries