fix: README typo; add Nvidia GPU troubleshooting

[Juliaj]

Main Changes

• Corrected a typo in README.md
• Added more details for troubleshooting O3DE start failure on Nvidia GPU machine.
• Fixed the issue that docker ps showed llama containers' as unhealthy. Health checks were failing because the llama containers’ default check hit port 8080, while the servers run on 8082 (embeddings) and 8083 (reranker). The fix is to add explicit health checks in docker/compose.*.yaml that curl the correct ports.

$ docker ps 
CONTAINER ID   IMAGE                                             COMMAND                  CREATED          STATUS                      PORTS     NAMES
...
0f4a4f8ebe7d   robotecai/mobile-manipulator-demo-llama:latest    "/app/llama-server -…"   19 minutes ago   Up 19 minutes (unhealthy)             docker-safety-reranker-1
f503cef71580   robotecai/mobile-manipulator-demo-llama:latest    "/app/llama-server -…"   19 minutes ago   Up 19 minutes (unhealthy)             docker-safety-embeddings-1

Testing
Following ran along with docker ps.

docker compose -f docker/compose.nvidia.yaml up