feat: add ```llm.requestDelay``` setting

[davidpissarra]

This PR introduces the setting llm.requestDelay. It represents the minimum time interval between requests. This setting avoids unnecessary API calls, typically when the user is still typing. For instance, if requestDelay is set to 2, the extension waits 2 seconds for something to be typed, otherwise makes an API call. Prior to this PR, every single character typed would result in a new call, overloading the API with many irrelevant requests, and most of the time making the user wait for all requests to be satisfied even if not needed anymore.

[davidpissarra]

An example of the current issue follows (different prompts with every typed character arriving at the rest API, which is cumbersome):

[McPatate]

Hello, thank you for opening this PR.

Two things:

• could you use the cancellation token passed in async provideInlineCompletionItems(document, position, context, token) { in your delay function?
• 1.5 secs is too high as a default value, could you set it to 150ms?

The plugin uses vscode.InlineCompletionItemProvider to handle the triggering of fetching completions. It already has a mechanism to buffer keypresses and delay requests, but it doesn't seem configurable and has a pretty low threshold. Additionally, requests are cancelled every time a new request is sent, in a remote endpoint setup it seems to work well, but I can see why in local setups that wouldn't, given there is no network overhead.

[pacman100]

This is definitely a great feature to have. For the time being, I am using a manual invocation shortcut to overcome this issue but this PR is a great way of resolving it.

[antoinejeannot]

Up, this really is a neat feature that will definitely help use local copilots. thanks @davidpissarra

[McPatate]

This was done in https://github.com/huggingface/llm-vscode/pull/111, closing