Improve error handling in AiProvider's AI command execution

AiProvider.execAiCmd()` currently has no real error handling — it always returns `true` regardless of what happens during AI task execution. This makes it impossible to detect and report failures from the AI assistant.

