New idea for handling intellisense

[rchiodo]

As I described here:
microsoft/vscode-jupyter#6333 (comment)

I had a new idea for handling intellisense that doesn't require a new pylance for every notebook (preliminary investigation seems to indicate about 100MB is allocated per instance).

The new idea is this:

• Python extension would skip providing intellisense for notebook and interactive cells. (DocumentSelector would just not include these)
• Jupyter extension would register a language server for every interpreter (after it's run at least once, or at least one for the active interpreter.
• Jupyter extension registration would use the same Document Selector for all servers. Namely notebook/interactive cells.
  • Normally this would mean duplicate intellisense (cause you would have multiple servers running that matched each notebook cell)
  • However the middleware addon (code I'm changing here) can swallow all intellisense requests when the request is for a notebook that doesn't match the interpreter associated with a notebook.
  • This is done by passing a 'shouldProvideIntellisense' function into the middleware
  • The jupyter extension will pass a different version of this function down to each middleware piece (basically something that binds itself to the interpreter in use and then compares the controller for a notebook URI against that interpreter)
  • This should then cause only 1 of the language server's results to be returned for a notebook.