Consider removing identifier orderings when doing identity resolution

[roji]

When loading related entities, we currently generate orderings to group together rows for the same entity. This is necessary for queries without identity resolution, since the shaper detects that new entities by the change in identifier. However, queries with identity resolution perform local lookup anyway, so these orderings aren't needed.

For example, consider a Blog with two collection navigations, Posts and Authors. When loading this entire graph, we currently generate:

SELECT [b].[Id], [b].[Name], [p].[Id], [p].[BlogId], [a].[Id], [a].[BlogId]
FROM [Blogs] AS [b]
LEFT JOIN [Post] AS [p] ON [b].[Id] = [p].[BlogId]
LEFT JOIN [Author] AS [a] ON [b].[Id] = [a].[BlogId]
ORDER BY [b].[Id], [p].[Id], [a].[Id]

Queries using identity resolution could omit the last two orderings.

This would potentially imply more local dictionary lookups, as rows are returned unordered and the shaper needs to jump back and forth between instances. My gut feeling is that the server-side ordering is much more expensive, but we'd have to confirm this first.

/cc @smitpatel