Experiment with threading AI into extraction refactorings

[DanielRosenwasser]

Today, TSServer provides several actions for "extract to function" and "extract to local" for both TypeScript and JavaScript. Currently, TSServer just creates a pre-generated name like newLocal or newFunction that's sufficiently unique. While this is fine, it would be interesting if we could use a language model to come up with a useful name and calculate a better name while TypeScript itself calculates the refactoring.

In cases where a language model can "confidently" calculate a new name, that name can replace the contents of the automatically-invoked rename for the refactor.

[spartanatreyu]

This should be controlled by a setting for sure.

Here are the outcomes I can see with this feature:

• A user refactors some code, likes the AI generated name and decides to use it.
• A user refactors some code and sees the AI generated name, but they want to make a slight tweak to the name.
• A user reactors some code and doesn't want to use the AI generated names at all.
  • Perhaps the user is consistently getting bad suggestions for the particular thing they are trying to do.
  • Perhaps seeing pre-generated names like newLocal or newFunction is a part of their proofing workflow (with eyeballs, linters, git hooks, or other code analysis tools).
  • The user is against using AI for some other reason

[zardoy]

I think TypeScript should try to infer name using common naming conventions at least (such as use... or get...), just like vscode-postfix-ts extension does:

From my experience it covers ~20% of code and can be useful if AI suggestions are slow, unavailable or a user doesn't want to use the AI generated names at all.