NO_SUCH_BUCKET and RESOURCE_UNAVAILABLE are not expected error types

[willdealtry]

We should not treat resource unavailable or no such bucket as expected S3 errors, as we might incorrectly decide that an object doesn't exist and can be newly created when in fact we've just not managed to talk to the storage to ascertain whether it's there or not.

[muhammadhamzasajjad]

We shouldn't remove RESOURCE_NOT_FOUND. As mentioned in this issue, RESOURCE_NOT_FOUND is returned when you call HeadObject() and the key is not found. You can also see this in failing tests in the CI. This came up during my exception normalization PRs (I guess I should've left a comment explaining why RESOURCE_NOT_FOUND is swallowed).

As for NO_SUCH_BUCKET, it is kept as per what is found here. Although, we might be able remove this with some more work.

[jamesmunro]

Please can we have a representative test? Something that shows the full issue and how this solves that.

Given the confusion about how these errors are used by AWS S3 SDK, I think it deserves a comment in the code as well please. Thanks.

Cannot merge due to Hamza's comments