notebook outputs api

[rebornix]

TODOs:

• Unify exthost and main thread output types
  • First step is reusing rich output for rendering always. Text and Error output now have special mimetypes. We will only have rich renderers (some defined in the core, some are contributed by extensions).
    • Legacy CellOutput and the new NotebookCellOutput are converted to IOutputDtoWithId before transferred to main thread.
    • CellOutputViewModel holds reference to the raw output and also the transformed one (stream to { 'application/x-notebook-stream': '' ) .
    • Renderers will only know about ITransformedDisplayOutputDto, which has outputId and [mimeType: string]: any.
  • Introduce NotebookCell.outputs2: NotebookCellOutput[]
    • We can probably do all the transform/addOutputId in the extension host, and then in the main thread we work with the rich output only ( { id: string;outputs: NotebookCellOutputItem[]; })
    • When we broadcast outputs changes to multiple extension host, we will need to create NotebookCellOutput objects in extension host.
  • Drop NotebookCell.outputs: ICellOutput[]
• Drop direct edit of metadata and outputs
  • Cell outputs
  • Cell metadata
  • Cell language
  • Document metadata
• Replace cell output data (mime type)
• Append cell output data (mime type)
• Adopt changes in diff editor
• Test
• Make sure all integration tests are enabled

[DonJayamanne]

@jrieken @roblourens
The current API might need some tweeking.

export class NotebookCellOutputItem {
    readonly metadata?: Record<string, string | number | boolean>;
}

The metadata only supports a flat keyvalue pair. However We don't control the metadata returned by Jupyter and it could (we don't know) contain complex data.
Please can we support a true dictionary (with support for nested items such as Record<string, any>)
As mentioned, we don't really control what metadata is returned & stored in these by Jupyter.