e.g., in #12480 (comment) we store duplicate cache entries for a wheel across multiple indexes because we do not eagerly compute hashes for all distributions and cannot store them in a content addressed key.
Since it's just a file, we can trivially hash them — we'll want to be careful about performance and consider the hashing algorithm though, like ideally we'd use Blake2 but the lockfile requires SHA-256 and it may not be worth hashing twice?