Replace MurmurHash3 with BLAKE3 in Hash module

[jasone]

Replace MurmurHash3 with BLAKE3 in the Hash module. With this change, hashing is no longer susceptible to pre-image attacks, so we don't need to randomly seed hashing anymore, and it may make sense to tweak the API to no longer refer to seed when removing the seeding logic.

BLAKE3 has some parallel hashing capabilities, so add something like hash_map in addition to hash_fold to expose this parallelism.

While performing this refactor, consider whether to idiomatically incorporate type ID into the hashes, as a mechanism for reducing brittleness when serializing composite data types.

[jasone]

In the long run, we may want to provide a function which can compute a hash for any immutable value. This would be a somewhat magical polymorphic function, in that it would use RTTI to traverse the value's transitive closure. However, the result would be pure, and it would commonly save hashing boilerplate in applications.

module Hash: sig
    module State: sig
        type t
        module Gen: sig
            val hash_a: 'a\& -> t -> t

This feature won't be worth the trouble in the bootstrap Basis though.

[jasone]

Right now Basis has to specially handle u128 in order to avoid cyclical dependencies. With the conversion to BLAKE3 this special handling would instead have to be for u256. However, we can refactor the Intw functors to enable generating a subset of the normal integer API that excludes Identifiable (and therefore hashing-related APIs). A call to the more limited Intw functor can be used instead of having a duplicate implementation of the relevant functions. Such a refactor should precede conversion to BLAKE3.