Improve AI security in GitLab with composite identities

Introduces the concept of 'composite identities' for AI agents, linking them to human user identities to enhance security and auditability. Details the design of composite identities, including how they are represented, authenticated, and authorized. Explains the redesign of the authorization framework to support multiple principals (AI agent and human user) for resource access. Mentions the enhancement of job tokens to signed JWTs and contributions to open-source libraries. Highlights the initial implementation of composite identities for service accounts with GitLab Duo with Amazon Q.