Refinements to prompt injection tests and benchmark

Summary:
- Refine tests to be more specific and accurate in some cases based on initial manual investigation of results.
- Use LLama-70b as judge model for best results
- Example results in the internal datasets folder
- Some logging and QOL changes.

Reviewed By: SimonWan

Differential Revision: D53904223

fbshipit-source-id: 2e2efff745d1f1ffb65640e3c35348f07861b953

CybersecurityBenchmarks/benchmark/llm.py

@@ -74,7 +74,7 @@ def _query_with_retries(
                 last_exception = exception
                 sleep_time = backoff_factor * (2**retry)
                 time.sleep(sleep_time)
-                LOG.warning(
+                LOG.debug(
                     f"LLM Query failed with error: {exception}. Sleeping for {sleep_time} seconds..."
                 )
         raise RuntimeError(