DNS: Spaghetti: Experimental node query builder

Note: This is an experiment representing my vision for how a
query builder should be architected and how it should look and feel.

- On-node configuration (no sidebar). This puts the node config right
  where the graph is and avoids mousing around, while also making it
  easier to read the graph as all the information is right there.
- Immutable (immer managed) node configuration allowing for simple
  and efficient undo/redo, and change detection.
- JS-side column name and type interence.
- JS-side materialization - no TP based serialization backend. Much
  simpler and just as effective + node highlighting while it's
  materializing.
- SQL folding, resulting in much more readable SQL code generation.
- Stable configs - when a node is detached from its parent, instead
  of forgetting its internal config its retains it and matches back
  columns via name when reconnected. If no such col name exists,
  highlight the column field in yellow for easy debugging.
- Node style:
  - Color coded titlebars with always visible names. This makes the
    graph much easier to interpret as there is a clear separation
    between the node title and its content.
  - Nodes are only draggable via their title bars, which allows for
    draggable elements within the body of the node.
  - Inputs and outputs are constrained to the left and right edges,
    resulting in more pleasing connection beziers.
- Very experimental LLM integration built in - allows asking LLMs to
  view and build graphs for you.

Change-Id: I6c93d9463982675e0e663e87fa05a9156907d1a7