CLI: llama-cli and llama-completion cosmetics

[andrew-aladev]

Hello, I've just finished the last (I think) PR regarding separation of llama-cli and llama-completion in docs and scripts. Here I tried to fix places where llama-cli and llama-completion are absolutely obvious, so it is safe to make updates.

1. Added llama-completion to .dockerignore
2. Added llama-completion to the GitHub issue template.
3. Fixed docs in copilot instructions.
4. Fixed writing a bash completion in README (because > implies non-interactive mode, so for now llama-cli is not possible here)
5. Used llama-completion in docs/backend/SYCL (llm_load_tensors: ... is the output of llama-completion)
6. Added llama-completion to docs/backend/hexagon/README
7. Used llama-completion in docs/backend/hexagon/developer (llama_model_loader: ... is the output of llama-completion)
8. Used llama-completion in scripts/fetch_server_test_models.py because it uses non-interactive mode.

PS I found other places where I can't understand whether llama-cli or llama-completion should be used. I think maintainers of these docs may update them later.

Related PRs: #17993, #18003

[andrew-aladev]

Thank you, I've resolved issues.