fix(aws): retry throttled paginated calls at the client layer

[bencehezso]

Summary

• paginate() in core/utils_aws.py retried a throttled page fetch by calling next() again on the same PageIterator. Since a PageIterator is a Python generator, once its body raises the frame is finished — every subsequent next() yields StopIteration, so the retry appeared to succeed and paginate() returned a truncated prefix as if it were the full result. Silent undercounts on exactly the accounts that get throttled most: large ones.
• Pushed retries down to botocore via AWS_RETRY_CONFIG = Config(retries={"mode": "adaptive", "max_attempts": 8}), applied on every session.client(...) in utils_aws.py and utils_egress_aws.py. Throttling is now retried at the HTTP layer inside the paginator's generator, where resuming is safe; paginate() becomes a plain for page in paginator.paginate(...) loop with no swallowing.
• Deleted the now-unused hand-rolled retry helpers (_retry_call, aws_api_call_with_retry) and their tests — keeping them alongside the client-level config invites the next contributor to reach for the wrong-layer pattern again.
• Added regression tests for paginate() / paginate_or_call(), including one that pins the no-swallowing contract by driving a paginator that raises ClientError(Throttling) mid-iteration and asserts the exception propagates instead of being silently truncated.