ras/serviceability: Refactor to use Podman utility with comprehensive VLLM testing

- Migrated from manual podman commands to avocado.utils.podman utility class
- Added setUp/tearDown methods for proper test lifecycle management
- Implemented four test methods for VFIO Spyre device access scenarios:
  * test_root_in_sentient: Root user with sentient group access (expects success)
  * test_root_not_in_sentient: Root user without sentient group (expects VFIO failure)
  * test_user_in_sentient: Non-root user with sentient group access (expects success)
  * test_user_not_in_sentient: Non-root user without sentient group (expects VFIO failure)
- Added wait_for_vllm_startup() helper for container readiness monitoring
- Created comprehensive YAML configuration with 4 variants:
  * RAG: 4 AIUs, 200G memory, granite-3.3-8b-instruct model
  * Entity Extraction: 1 AIU, 100G memory
  * Re-Ranker: 1 AIU, 50G memory, bge-reranker-large model
  * Embedding: 1 AIU, 50G memory, granite-embedding-125m-english model
- All container runtime parameters now configurable via YAML:
  * Memory, shared memory, device, privileged mode, PIDs limit
  * User namespace, group add, port mapping
  * VLLM-specific options: prefix caching, additional arguments
- Removed hardcoded credentials for security
- Added comprehensive inline documentation in YAML
- Registry login with API key support
- Container image pull before test execution

Signed-off-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>

Author: abdhaleegit