feat: top level ordering

[bojan88]

Support for top level ordering using dot notation. https://postgrest.org/en/v12/references/api/resource_embedding.html#top-level-ordering

No interface changes. The only change is supported fields for column in .order([{ column: 'column_name' }]). Instead of only accepting current table column names, it now accepts embedded fields with dot notation to drill down.

Example:

const query = pgClient
  .query('films')
  .select('title')
  .select(pgClient.embeddedQuery('directors', 'one').select('*'))
  .order([{ column: 'directors.last_name', order: 'desc' }]);

---

Alternative considerations

We considered different approaches to implement this and current implementation was more difficult to implement but better for developer experience.

One of the alternatives we considered: Using more properties for order method parameters to say it's a top level query (top: true). Example:

const query = pgClient
  .query('films')
  .select('title')
  .select(pgClient.embeddedQuery('directors', 'one')
    .select('*')
    .order([{ column: 'last_name', order: 'desc', top: true }])
  );

The problem with this approach was with order precedence when top level and root orders are used together. We had problems constructing a query with both of these options order=id,films(id) and order=films(id),id. We considered 2 solutions - precedence based on method call order, and one more property for precedence like weight: number. Both of these options looked like unintuitive, so instead we ended up with dot notation on root query (implementation in this PR)

---

Note that documentation will be added later after this is released.

[kegjon]

I like the interface. The dot-notation for referencing embedded fields I think is fairly clear.

My only request was going to be to add the docs, but given you intend to already, I have nothing to add. Looks good.