docs: add SkyPilot Kubernetes tutorial

[zeel2104]

What does this PR do ?

Adds a beginner-friendly SkyPilot + Kubernetes tutorial for NeMo AutoModel, including ready-to-run single-node and two-node example configs, and fixes SkyPilot launcher env-var interpolation so ${HF_TOKEN} works as documented.

Changelog

Add docs/launcher/skypilot-kubernetes.md with a step-by-step SkyPilot + Kubernetes tutorial
Add single-node example config at examples/llm_finetune/llama3_2/llama3_2_1b_squad_skypilot_kubernetes.yaml
Add two-node example config at examples/llm_finetune/llama3_2/llama3_2_1b_squad_skypilot_kubernetes_2nodes.yaml
Link the new tutorial from docs/launcher/overview.md, docs/launcher/skypilot.md, and docs/index.md
Resolve SkyPilot launcher env vars in nemo_automodel/cli/app.py
Add unit test coverage for SkyPilot env-var resolution in tests/unit_tests/_cli/test_app.py

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?

-Targeted SkyPilot launcher tests passed locally
-automodel --help works locally
-Real SkyPilot + Kubernetes validation could not be completed from native Windows because SkyPilot requires the -Unix resource module
-Hardware validation is still needed on Linux

Additional Information

• Related to Skypilot + k8s tutorial #1663

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[akoumpa]

Hi @zeel2104 , thank you writing this up, for testing this myself, can you share the provider you used for testing this?

Thank you.

[zeel2104]

Hi @akoumpa, I haven’t validated this on a provider yet. Locally, I did verify the related SkyPilot launcher path by running the targeted unit tests:

tests/unit_tests/_cli/test_app.py
tests/unit_tests/_cli/test_skypilot_app.py
tests/unit_tests/launcher/test_skypilot_config.py
tests/unit_tests/launcher/test_skypilot_utils.py
These passed locally, and automodel --help also works after reinstalling the package in my environment.

I can validate the SkyPilot CLI path from a Linux VM, including sky startup, local launcher behavior, and Kubernetes client-side checks, but full end-to-end validation of the tutorial still requires access to a real GPU Kubernetes cluster/provider.
For the same , do we have an internal testbed, shared Kubernetes cluster, or cloud credits/access for something like GKE/EKS/AKS that contributors can use to validate SkyPilot + Kubernetes workflows?

[akoumpa]

/ok to test 1fca216

[jgerh]

Completed a tech pubs review of .md files and added a few copyedits and style adjustments to align with our style guide.

[akoumpa]

/ok to test 0f8cad1

[zeel2104]

@jgerh
Addressed in follow-up commits. I applied the requested copyedits in docs/launcher/overview.md and docs/launcher/skypilot.md.

[akoumpa]

Thank you @zeel2104 looks good 🙇