[Feature Request] Implement configurable trigger debounce for local inference

[h8f1z]

Background
I am running llama-vscode on a CPU-only system. While the extension is excellent, it triggers a completion request almost immediately upon typing.

The Problem
The extension still initiates the local LLM backend immediately. On low-end hardware, this causes:

1. Input Latency: Typing becomes laggy as the CPU spikes to handle the inference request. CPU is at 100% whenever I type (also loud).
2. Resource Waste: Multiple requests are sent for "half-finished" thoughts, which are immediately cancelled/discarded as I continue typing.

Proposed Solution

• Add an extension-specific setting, e.g., llama.autocomplete.debounceDelay (default 0ms).
• When set (e.g., to 3000ms), the extension should wait for a period of "idle" time after the last keystroke before calling the backend.

This would ensure that only "intentional" pauses in typing trigger the CPU-intensive local model.

Alternatives Considered

• Manual Triggering: Possible, but a debounce offers a much smoother "automatic" experience for local users. No extra keystrokes (which is taken by another shortcut = another problem).

I am happy to contribute a PR for this if the maintainers agree with the approach.

[igardev]

@h8f1z There is a manual mode - set the setting "Auto" to false (uncheck it). In this case the triggering will be only manual (Ctrl+L). For the debounce idea - I wil think about it. In general I prefer to keep it as simple as possible.

[h8f1z]

@h8f1z There is a manual mode - set the setting "Auto" to false (uncheck it). In this case the triggering will be only manual (Ctrl+L). For the debounce idea - I wil think about it. In general I prefer to keep it as simple as possible.

I did try the manual mode. But it didn't quite work. I prefer llama.vscode for it's simplicity, and I wouln't want to change it. That said, I thought that a debounce feature could improve the efficiency of the extension, and save at least some of the heat generated by the PC. Making it configuration was mentioned because some of the users may not like a fixed interval.

Thank you for considering and the contribution.

[igardev]

@h8f1z OK, please create a pull request - I will have a look.