fix: fix hashing specialization

[ritchie46]

fixes #14749

[nameexhaustion]

Some other places (all?), (e.g. numeric_vec_hash_combine) that previously hashed &T were also changed by #14648 to hash T instead of &T, but I think those are ok since they aren't mixing between hashes of T and &T. group_by_threaded_slice/iter were problematic because they used insert_with_hasher that allowed us to insert a hashing function that did not hash the same way as the initial key.

[ritchie46]

Some other places (all?)

I believe not, but please double check. The more eyes over this the better.

[nameexhaustion]

I'm now actually inclined to believe hash_one(x) should have been equivalent to hash_one(&x) - opened an issue with aHash at tkaitchuck/aHash#214