It would be nice to have a functional version of hash-table-union! with this signature:

  (hash-table-union hash-table₁ hash-table₂ merge-proc)

Rather than mutating either input hash table, it would return a new one of the same type that contained the union of the keys.  Each key would be associated with a value returned by merge-proc when given the values at that key in the input hash tables.  #f would be passed as input when one hash table didn't contain that key.

In fact, the imperative version should take an optional merge-proc, too, rather than just overwriting the second hash table's values.  That would make it handle many more use cases.