Add focused NeMo Retriever agent skills

[jeremyjordan]

Description

This PR replaces the current monolithic Retriever skill with five focused skills: setup, ingest, query, evaluate, and service. The skills are packaged under nemo_retriever/.agents/skills, exposed through per-skill symlinks in .agents/skills and .claude/skills, and included in the wheel package data.

It also updates skill evaluation support so split skill directories and symlinked skills can be copied into isolated evaluation workspaces. Legacy slash-command aliases now route to the task-specific Retriever skills.

While validating the local workflows, this surfaced a gap in the local-HF query path: ingest and recall supported HuggingFace local embedding controls, but retriever query and retriever eval export did not expose equivalent flags.
This PR adds those local query embedding options and covers them with focused tests. I'm including the changes in this PR because the skills depend on them, but they should likely be separated out into a different focused MR before merging.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• If adjusting docker-compose.yaml environment variables have you ensured those are mimicked in the Helm values.yaml file.

[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.