docs: note CUDA container and hf build requirements

[tryingET]

This docs-only PR adds two CUDA build notes that came up while bootstrapping a clean Blackwell Gemma 4 canary:

1. if you build inside an nvidia/cuda:* container, the build container also needs --gpus all, otherwise the final link step can fail with missing libcuda.so.1 / unresolved CUDA driver symbols
2. if you want to use -hf / --hf-repo, you need OpenSSL development headers available at configure time (for example libssl-dev + pkg-config on Debian/Ubuntu-based systems), otherwise llama-server -hf ... can fail at runtime with the existing HTTPS is not supported message

The PR keeps this as one change because both notes belong to the same user-facing build path in docs/getting-started.md, and adds one concrete CUDA container example that includes the required packages/flags.