Add expression based cursor

[nikolay-slavov]

Hi, this merge request adds an abilty to build expression cursor.
When we need to paginate records and order them by value generated on the fly,
we will need that value in the both order by and where clauses.

Use case:

If we need to paginate records with cursor_fields: [:rank_value, :id], where
rank_value is calculated on the fly we can do the following:

  query =
    from(
      f in Post,
      select_merge: %{
        rank_value:
          fragment("ts_rank(document, plainto_tsquery('simple', ?)) AS rank_value", ^q)
      },
      where: fragment("document @@ plainto_tsquery('simple', ?)", ^q),
      order_by: [
        desc: fragment("rank_value"),
        desc: f.id
      ]
    )

    query
    |> Repo.paginate(
      limit: 30,
      fetch_cursor_value_fun: fn
        # Here we build the rank_value for each returned row
        schema, {:rank_value, _} ->
          {:ok, %{rows: [[rank_value]]}} =
            Repo.query("SELECT ts_rank($1, plainto_tsquery('simple', $2))", [
              schema.document,
              q
            ])

          rank_value

        schema, field ->
          Paginator.default_fetch_cursor_value(schema, field)
      end,
      cursor_fields: [
        {:rank_value, # Here we build the rank_value that will be used in the where clause
         fn ->
           dynamic(
             [x],
             fragment("ts_rank(document, plainto_tsquery('simple', ?))", ^q)
           )
         end},
        :id
      ]
    )

From example above we can see that the cursor_fields now accepts and tuples.
The first element of the tuple is the name of the cursor field and the second one is a
function responsible for how the value should be generated.

[quexpl]

Thanks @nikolay-slavov great job! I'm sorting by date field which is inside json map, at this point I'm not using embedded_schema so your PR was a bless for me. Unfortunately I had few issues after pulling it to most recent version from main branch.
After changes in #68 cursor is now a map rather than list which impacts your changes.
Now cursor is a map for example: %{rank_value: customer_3.rank_value, id: customer_3.id}. When an expression-based cursor is used, the entire tuple of the cursor is used as the field name, which caused decoding errors:

** (ArgumentError) cannot deserialize #Function<20.33583418/0 in Webapp.Machines.page_machines/3>, the term is not safe for deserialization

[telphan]

Closing in favour of #166