GPU-AV: Find ways to prevent bad commands / pipelines from executing

For pipelines, we can try to add a `if (gpu_av_kill_pipeline_execution) { return; }` to every entry point using instrumentation.
For commands, no good answer, it will probably be case by case solutions.

The issue names GPU-AV pipelines, commands, instrumentation, and the gpu_av_kill_pipeline_execution flag, but no files or tests. Start by locating the instrumentation entry points and existing kill-flag handling. Define the expected behavior for pipeline and command execution, then add coverage showing that bad work does not execute.

Written by the indexing model from the issue text.