Soften stance on AI-generated bug reports

The language about never submitting issues with AI is a bit hard-line. Not all AI-help with issues is bad, can probably soften it a little. The main point is not taking LLM assurances about bugs at face value without being absolutely sure. Then also respecting the upstream repo policies / issue templates etc. for reporting, and supplying a minimal replication.