ai-red-teaming: add hands-on exercises to three topics

[DmitriyKozyatinskiy]

Adds one @course@ link to three topics where the existing resources are articles and papers, so a learner can run the attack rather than only read about it.

Topic	Links before	Added
Prompt Injection	5	Inject instructions into a live assistant's context and observe the override, then see the mitigation
Agentic AI Security	4	Hijack an agent's goal mid-task, matching the goal-hijacking risk the topic describes
LLM Security Testing	3	The full OWASP LLM, Agentic and MCP Top 10 track, one exploit-then-fix exercise per risk

I left Jailbreak Techniques alone because we have no exercise that genuinely matches it, and I would rather not stretch a link to fit.

Disclosure: I work on RansomLeak, which hosts these exercises. They are free for individual learners with no sign-up. Same format as #10287, which was merged into the cyber-security roadmap.

https://claude.ai/code/session_01N4UCQWsmoH6URE6YjxPWgX

[DmitriyKozyatinskiy]

Consolidating into #10297 so this is a single PR, per the "Create a Single PR for Content Additions" guideline. Same content, nothing dropped. Sorry for the churn.