- in-mem ignored None keys
- cassandra yielded input dicts as-is instead of a dict with just the hashes
Looks fine save from a few comments/questions.
The "content key algortithm" stuff should probably be moved to a helper in swh.model, as we keep repeating it everywhere.
Took me a while to understand this logic. Why not use _content_key_algorithm() here?
You could also make a dict with these keys, which would alleviate the n² nature of the lookup (even though it's probably not a big deal considering how tiny n is)
As you only iterate once, no need for the list(content)