Prompt Injection & LLM-misuse scanner

New specialized scanner that analyzes application code calling LLMs (with auto-detection for Anthropic, OpenAI, Bedrock, and Vertex SDKs) to identify: non-parameterized prompts (user input concatenation), lack of guardrails (no system prompt enforcement), indirect prompt injection (unvalidated tool output reinjection), unsafe tool use (file/exec/network access without allowlists), lack of rate/cost-limiting, and secret leaks within prompts according to OWASP LLM Top 10 patterns.