docs: Update spec to reflect Helm-based deployment and Trainium2 support

- Add Trainium2 requirements (Req 12) and model compilation requirement
- Add Helm chart deployment requirements (Req 11)
- Update model memory requirements (Req 13) for Scout/Maverick
- Update glossary with new terms (Helm, Trainium2, NxD Inference, etc.)
- Update tasks to reflect Helm-based workflow
- Add notes about FP8 quantization for GPU vs BF16 for Trainium2

Author: devfloor9