Security Vulnerability Detection with Multitask Self-Instructed Fine-Tuning of Large Language Models