remember and avoid URLs that failed for non-retryable reasons

Some NFT collections specify their metadata URIs in a wrong way e.g. by returning a URL for an index of all the JSON files for every token (for example https://freepunks.xyz/json/) instead of a URL to the specific token file.

This causes the service to respond correctly with errors like `MetadataSizeExceededError`. However, since a collection can have thousands of tokens, this URL fetch is performed thousands of times all with the same result.

We could figure out a way to remember these faulty URLs and automatically mark token jobs as failed without fetching anything if they use some of these incorrect URLs again.